external
/
citus
mirror of https://github.com/citusdata/citus.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/master' into columnar-index 

Conflicts:
	src/test/regress/expected/columnar_empty.out
	src/test/regress/expected/multi_extension.out
pull/5052/head
Onur Tirtir 2021-06-16 20:23:50 +03:00
parent a2efe59e2f 27c7d28f7f
commit 3d11c0f9ef
341 changed files with 10513 additions and 3616 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
.circleci/config.yml
View File

@ -64,6 +64,12 @@ jobs:
- run:
name: 'Check if changed'
command: git diff --exit-code
- run:
name: 'Check for gitignore entries .for source files'
command: ci/fix_gitignore.sh
- run:
name: 'Check if changed'
command: git diff --exit-code
- run:
name: 'Check for lengths of changelog entries'
command: ci/disallow_long_changelog_entries.sh
@ -445,6 +451,12 @@ workflows:
image_tag: '12.6'
make: check-multi
requires: [build-12]
- test-citus:
name: 'test-12_check-multi-1'
pg_major: 12
image_tag: '12.6'
make: check-multi-1
requires: [build-12]
- test-citus:
name: 'test-12_check-mx'
pg_major: 12
@ -513,6 +525,12 @@ workflows:
image_tag: '13.2'
make: check-multi
requires: [build-13]
- test-citus:
name: 'test-13_check-multi-1'
pg_major: 13
image_tag: '13.2'
make: check-multi-1
requires: [build-13]
- test-citus:
name: 'test-13_check-mx'
pg_major: 13

84
CHANGELOG.md
View File

@ -1,3 +1,87 @@
### citus v10.1.0 (June 15, 2021) ###
* Drops support for PostgreSQL 11
* Adds `shard_count` parameter to `create_distributed_table` function
* Adds support for `ALTER DATABASE OWNER`
* Adds support for temporary columnar tables
* Adds support for using sequences as column default values when syncing
metadata
* `alter_columnar_table_set` enforces columnar table option constraints
* Continues to remove shards after failure in `DropMarkedShards`
* Deprecates the `citus.replication_model` GUC
* Enables `citus.defer_drop_after_shard_move` by default
* Ensures free disk space before moving a shard
* Fetches shard size on the fly for the rebalance monitor
* Ignores old placements when disabling or removing a node
* Implements `improvement_threshold` at shard rebalancer moves
* Improves orphaned shard cleanup logic
* Improves performance of `citus_shards`
* Introduces `citus.local_hostname` GUC for connections to the current node
* Makes sure connection is closed after each shard move
* Makes sure that target node in shard moves is eligible for shard move
* Optimizes partitioned disk size calculation for shard rebalancer
* Prevents connection errors by properly terminating connections
* Prevents inheriting a distributed table
* Prevents users from dropping & truncating known shards
* Pushes down `VALUES` clause as long as not in outer part of a `JOIN`
* Reduces memory usage for multi-row inserts
* Reduces memory usage while rebalancing shards
* Removes length limits around partition names
* Executor avoids opening extra connections
* Fixes a bug that can cause a crash when DEBUG4 logging is enabled
* Fixes data race in `get_rebalance_progress`
* Fixes error message for local table joins
* Fixes `FROM ONLY` queries on partitioned tables
* Fixes issues caused by omitting public schema in queries
* Fixes nested `SELECT` query with `UNION` bug
* Fixes null relationName bug at parallel execution
* Fixes possible segfaults when using Citus in the middle of an upgrade
* Fixes problems with concurrent calls of `DropMarkedShards`
* Fixes shared dependencies that are not resident in a database
* Fixes stale hostnames bug in prepared statements after `master_update_node`
* Fixes using 2PC when it might be necessary
* Preserves access method of materialized views when undistributing
or altering distributed tables
### citus v8.3.3 (March 23, 2021) ###
* Fixes a bug that leads to various issues when a connection is lost

2
README.md
View File

@ -2,7 +2,7 @@
[![Slack Status](http://slack.citusdata.com/badge.svg)](https://slack.citusdata.com)
[![Latest Docs](https://img.shields.io/badge/docs-latest-brightgreen.svg)](https://docs.citusdata.com/)
[![Code Coverage](https://codecov.io/gh/citusdata/citus/branch/master/graph/badge.svg)](https://codecov.io/gh/citusdata/citus/branch/master/graph/badge.svg)
[![Code Coverage](https://codecov.io/gh/citusdata/citus/branch/master/graph/badge.svg)](https://app.codecov.io/gh/citusdata/citus)
## What is Citus?

16
ci/README.md
View File

@ -156,9 +156,9 @@ git merge "community/$PR_BRANCH"
familiar with the change.
5. You should rerun the `check-merge-to-enterprise` check on
`community/$PR_BRANCH`. You can use re-run from failed option in circle CI.
6. You can now merge the PR on community. Be sure to NOT use "squash and merge",
6. You can now merge the PR on enterprise. Be sure to NOT use "squash and merge",
but instead use the regular "merge commit" mode.
7. You can now merge the PR on enterprise. Be sure to NOT use "squash and merge",
7. You can now merge the PR on community. Be sure to NOT use "squash and merge",
but instead use the regular "merge commit" mode.
The subsequent PRs on community will be able to pass the
@ -346,3 +346,15 @@ foo = 2
#endif
```
This was deemed to be error prone and not worth the effort.
## `fix_gitignore.sh`
This script checks and fixes issues with `.gitignore` rules:
1. Makes sure git ignores the `.sql` files and expected output files that are generated
from `.source` template files. If you created or deleted a `.source` file in a commit,
git ignore rules should be updated to reflect this change.
2. Makes sure we do not commit any generated files that should be ignored. If there is an
ignored file in the git tree, the user is expected to review the files that are removed
from the git tree and commit them.

8
ci/check_enterprise_merge.sh
View File

@ -65,6 +65,14 @@ fi
# undo partial merge
git merge --abort
# If we have a conflict on enterprise merge on the master branch, we have a problem.
# Provide an error message to indicate that enterprise merge is needed.
if [[ $PR_BRANCH = master ]]; then
echo "ERROR: Master branch has merge conlicts with enterprise-master."
echo "Try re-running this job if you merged community PR before enterprise PR. Otherwise conflicts need to be resolved as a separate PR on enterprise."
exit 1
fi
if ! git fetch enterprise "$PR_BRANCH" ; then
echo "ERROR: enterprise/$PR_BRANCH was not found and community PR branch could not be merged into enterprise-master"
exit 1

34
ci/fix_gitignore.sh Executable file
View File

@ -0,0 +1,34 @@
#! /bin/bash
# shellcheck disable=SC2012
set -euo pipefail
# shellcheck disable=SC1091
source ci/ci_helpers.sh
# We list all the .source files in alphabetical order, and do a substitution
# before writing the resulting file names that are created by those templates in
# relevant .gitignore files
#
# 1. Capture the file name without the .source extension
# 2. Add the desired extension at the end
# 3. Add a / character at the beginning of each line to conform to .gitignore file format
#
# e.g. multi_copy.source -> /multi_copy.sql
ls -1 src/test/regress/input | sed -E "s#(.*)\.source#/\1.sql#" > src/test/regress/sql/.gitignore
# e.g. multi_copy.source -> /multi_copy.out
ls -1 src/test/regress/output | sed -E "s#(.*)\.source#/\1.out#" > src/test/regress/expected/.gitignore
# Remove all the ignored files from git tree, and error out
# find all ignored files in git tree, and use quotation marks to prevent word splitting on filenames with spaces in them
ignored_lines_in_git_tree=$(git ls-files --ignored --exclude-standard | sed 's/.*/"&"/')
if [[ -n $ignored_lines_in_git_tree ]]
then
echo "Ignored files should not be in git tree!"
echo "${ignored_lines_in_git_tree}"
echo "Removing these files from git tree, please review and commit"
echo "$ignored_lines_in_git_tree" | xargs git rm -r --cached
exit 1
fi

1
ci/fix_style.sh
View File

@ -14,3 +14,4 @@ ci/remove_useless_declarations.sh
ci/disallow_c_comments_in_migrations.sh
ci/disallow_long_changelog_entries.sh
ci/normalize_expected.sh
ci/fix_gitignore.sh

18
configure vendored
View File

@ -1,6 +1,6 @@
#! /bin/sh
# Guess values for system-dependent variables and create Makefiles.
# Generated by GNU Autoconf 2.69 for Citus 10.1devel.
# Generated by GNU Autoconf 2.69 for Citus 10.2devel.
#
#
# Copyright (C) 1992-1996, 1998-2012 Free Software Foundation, Inc.
@ -579,8 +579,8 @@ MAKEFLAGS=
# Identity of this package.
PACKAGE_NAME='Citus'
PACKAGE_TARNAME='citus'
PACKAGE_VERSION='10.1devel'
PACKAGE_STRING='Citus 10.1devel'
PACKAGE_VERSION='10.2devel'
PACKAGE_STRING='Citus 10.2devel'
PACKAGE_BUGREPORT=''
PACKAGE_URL=''
@ -1260,7 +1260,7 @@ if test "$ac_init_help" = "long"; then
# Omit some internal or obsolete options to make the list less imposing.
# This message is too long to be a string in the A/UX 3.1 sh.
cat <<_ACEOF
\`configure' configures Citus 10.1devel to adapt to many kinds of systems.
\`configure' configures Citus 10.2devel to adapt to many kinds of systems.
Usage: $0 [OPTION]... [VAR=VALUE]...
@ -1322,7 +1322,7 @@ fi
if test -n "$ac_init_help"; then
case $ac_init_help in
short | recursive ) echo "Configuration of Citus 10.1devel:";;
short | recursive ) echo "Configuration of Citus 10.2devel:";;
esac
cat <<\_ACEOF
@ -1425,7 +1425,7 @@ fi
test -n "$ac_init_help" && exit $ac_status
if $ac_init_version; then
cat <<\_ACEOF
Citus configure 10.1devel
Citus configure 10.2devel
generated by GNU Autoconf 2.69
Copyright (C) 2012 Free Software Foundation, Inc.
@ -1908,7 +1908,7 @@ cat >config.log <<_ACEOF
This file contains any messages produced by compilers while
running configure, to aid debugging if configure makes a mistake.
It was created by Citus $as_me 10.1devel, which was
It was created by Citus $as_me 10.2devel, which was
generated by GNU Autoconf 2.69. Invocation command line was
$ $0 $@
@ -5356,7 +5356,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
# report actual input values of CONFIG_FILES etc. instead of their
# values after options handling.
ac_log="
This file was extended by Citus $as_me 10.1devel, which was
This file was extended by Citus $as_me 10.2devel, which was
generated by GNU Autoconf 2.69. Invocation command line was
CONFIG_FILES = $CONFIG_FILES
@ -5418,7 +5418,7 @@ _ACEOF
cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
ac_cs_config="`$as_echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`"
ac_cs_version="\\
Citus config.status 10.1devel
Citus config.status 10.2devel
configured by $0, generated by GNU Autoconf 2.69,
with options \\"\$ac_cs_config\\"

2
configure.in
View File

@ -5,7 +5,7 @@
# everyone needing autoconf installed, the resulting files are checked
# into the SCM.
AC_INIT([Citus], [10.1devel])
AC_INIT([Citus], [10.2devel])
AC_COPYRIGHT([Copyright (c) Citus Data, Inc.])
# we'll need sed and awk for some of the version commands

18
src/backend/columnar/columnar_tableam.c
View File

@ -2076,6 +2076,15 @@ alter_columnar_table_set(PG_FUNCTION_ARGS)
if (!PG_ARGISNULL(1))
{
options.chunkRowCount = PG_GETARG_INT32(1);
if (options.chunkRowCount < CHUNK_ROW_COUNT_MINIMUM ||
options.chunkRowCount > CHUNK_ROW_COUNT_MAXIMUM)
{
ereport(ERROR, (errmsg("chunk group row count limit out of range"),
errhint("chunk group row count limit must be between "
UINT64_FORMAT " and " UINT64_FORMAT,
(uint64) CHUNK_ROW_COUNT_MINIMUM,
(uint64) CHUNK_ROW_COUNT_MAXIMUM)));
}
ereport(DEBUG1,
(errmsg("updating chunk row count to %d", options.chunkRowCount)));
}
@ -2084,6 +2093,15 @@ alter_columnar_table_set(PG_FUNCTION_ARGS)
if (!PG_ARGISNULL(2))
{
options.stripeRowCount = PG_GETARG_INT32(2);
if (options.stripeRowCount < STRIPE_ROW_COUNT_MINIMUM ||
options.stripeRowCount > STRIPE_ROW_COUNT_MAXIMUM)
{
ereport(ERROR, (errmsg("stripe row count limit out of range"),
errhint("stripe row count limit must be between "
UINT64_FORMAT " and " UINT64_FORMAT,
(uint64) STRIPE_ROW_COUNT_MINIMUM,
(uint64) STRIPE_ROW_COUNT_MAXIMUM)));
}
ereport(DEBUG1, (errmsg(
"updating stripe row count to " UINT64_FORMAT,
options.stripeRowCount)));

2
src/backend/distributed/citus.control
View File

@ -1,6 +1,6 @@
# Citus extension
comment = 'Citus distributed database'
default_version = '10.1-1'
default_version = '10.2-1'
module_pathname = '$libdir/citus'
relocatable = false
schema = pg_catalog

19
src/backend/distributed/commands/alter_table.c
View File

@ -204,7 +204,6 @@ static char * GetAccessMethodForMatViewIfExists(Oid viewOid);
static bool WillRecreateForeignKeyToReferenceTable(Oid relationId,
CascadeToColocatedOption cascadeOption);
static void WarningsForDroppingForeignKeysWithDistributedTables(Oid relationId);
static void ExecuteQueryViaSPI(char *query, int SPIOK);
PG_FUNCTION_INFO_V1(undistribute_table);
PG_FUNCTION_INFO_V1(alter_distributed_table);
@ -219,11 +218,11 @@ PG_FUNCTION_INFO_V1(worker_change_sequence_dependency);
Datum
undistribute_table(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
bool cascadeViaForeignKeys = PG_GETARG_BOOL(1);
CheckCitusVersion(ERROR);
TableConversionParameters params = {
.relationId = relationId,
.cascadeViaForeignKeys = cascadeViaForeignKeys
@ -243,6 +242,8 @@ undistribute_table(PG_FUNCTION_ARGS)
Datum
alter_distributed_table(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
char *distributionColumn = NULL;
@ -280,9 +281,6 @@ alter_distributed_table(PG_FUNCTION_ARGS)
}
}
CheckCitusVersion(ERROR);
TableConversionParameters params = {
.relationId = relationId,
.distributionColumn = distributionColumn,
@ -305,13 +303,13 @@ alter_distributed_table(PG_FUNCTION_ARGS)
Datum
alter_table_set_access_method(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *accessMethodText = PG_GETARG_TEXT_P(1);
char *accessMethod = text_to_cstring(accessMethodText);
CheckCitusVersion(ERROR);
TableConversionParameters params = {
.relationId = relationId,
.accessMethod = accessMethod
@ -558,8 +556,11 @@ ConvertTable(TableConversionState *con)
includeIndexes = false;
}
bool includeReplicaIdentity = true;
List *postLoadCommands = GetPostLoadTableCreationCommands(con->relationId,
includeIndexes);
includeIndexes,
includeReplicaIdentity);
List *justBeforeDropCommands = NIL;
List *attachPartitionCommands = NIL;

6
src/backend/distributed/commands/citus_add_local_table_to_metadata.c
View File

@ -179,7 +179,7 @@ remove_local_tables_from_metadata(PG_FUNCTION_ARGS)
* properties:
* - it will have only one shard,
* - its distribution method will be DISTRIBUTE_BY_NONE,
* - its replication model will be ReplicationModel,
* - its replication model will be REPLICATION_MODEL_STREAMING,
* - its replication factor will be set to 1.
* Similar to reference tables, it has only 1 placement. In addition to that, that
* single placement is only allowed to be on the coordinator.
@ -996,9 +996,7 @@ InsertMetadataForCitusLocalTable(Oid citusLocalTableId, uint64 shardId)
Assert(shardId != INVALID_SHARD_ID);
char distributionMethod = DISTRIBUTE_BY_NONE;
char replicationModel = ReplicationModel;
Assert(replicationModel != REPLICATION_MODEL_2PC);
char replicationModel = REPLICATION_MODEL_STREAMING;
uint32 colocationId = INVALID_COLOCATION_ID;
Var *distributionColumn = NULL;

127
src/backend/distributed/commands/create_distributed_table.c
View File

@ -86,13 +86,9 @@
*/
#define LOG_PER_TUPLE_AMOUNT 1000000
/* Replication model to use when creating distributed tables */
int ReplicationModel = REPLICATION_MODEL_COORDINATOR;
/* local function forward declarations */
static char DecideReplicationModel(char distributionMethod, bool viaDeprecatedAPI);
static char DecideReplicationModel(char distributionMethod, char *colocateWithTableName,
bool viaDeprecatedAPI);
static void CreateHashDistributedTableShards(Oid relationId, int shardCount,
Oid colocatedTableId, bool localTableEmpty);
static uint32 ColocationIdForNewTable(Oid relationId, Var *distributionColumn,
@ -119,7 +115,6 @@ static List * GetFKeyCreationCommandsRelationInvolvedWithTableType(Oid relationI
int tableTypeFlag);
static Oid DropFKeysAndUndistributeTable(Oid relationId);
static void DropFKeysRelationInvolvedWithTableType(Oid relationId, int tableTypeFlag);
static bool LocalTableEmpty(Oid tableId);
static void CopyLocalDataIntoShards(Oid relationId);
static List * TupleDescColumnNameList(TupleDesc tupleDescriptor);
static bool DistributionColumnUsesGeneratedStoredColumn(TupleDesc relationDesc,
@ -146,12 +141,11 @@ PG_FUNCTION_INFO_V1(create_reference_table);
Datum
master_create_distributed_table(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *distributionColumnText = PG_GETARG_TEXT_P(1);
Oid distributionMethodOid = PG_GETARG_OID(2);
CheckCitusVersion(ERROR);
EnsureCitusTableCanBeCreated(relationId);
char *colocateWithTableName = NULL;
@ -193,6 +187,8 @@ master_create_distributed_table(PG_FUNCTION_ARGS)
Datum
create_distributed_table(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
if (PG_ARGISNULL(0) || PG_ARGISNULL(1) || PG_ARGISNULL(2) || PG_ARGISNULL(3))
{
PG_RETURN_VOID();
@ -225,8 +221,6 @@ create_distributed_table(PG_FUNCTION_ARGS)
shardCountIsStrict = true;
}
CheckCitusVersion(ERROR);
EnsureCitusTableCanBeCreated(relationId);
/* enable create_distributed_table on an empty node */
@ -275,6 +269,7 @@ create_distributed_table(PG_FUNCTION_ARGS)
Datum
create_reference_table(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
char *colocateWithTableName = NULL;
@ -282,8 +277,6 @@ create_reference_table(PG_FUNCTION_ARGS)
bool viaDeprecatedAPI = false;
CheckCitusVersion(ERROR);
EnsureCitusTableCanBeCreated(relationId);
/* enable create_reference_table on an empty node */
@ -442,6 +435,7 @@ CreateDistributedTable(Oid relationId, Var *distributionColumn, char distributio
EnsureDependenciesExistOnAllNodes(&tableAddress);
char replicationModel = DecideReplicationModel(distributionMethod,
colocateWithTableName,
viaDeprecatedAPI);
/*
@ -464,7 +458,7 @@ CreateDistributedTable(Oid relationId, Var *distributionColumn, char distributio
EnsureReferenceTablesExistOnAllNodes();
/* we need to calculate these variables before creating distributed metadata */
bool localTableEmpty = LocalTableEmpty(relationId);
bool localTableEmpty = TableEmpty(relationId);
Oid colocatedTableId = ColocatedTableId(colocationId);
/* create an entry for distributed table in pg_dist_partition */
@ -631,44 +625,38 @@ DropFKeysRelationInvolvedWithTableType(Oid relationId, int tableTypeFlag)
/*
* DecideReplicationModel function decides which replication model should be
* used depending on given distribution configuration and global ReplicationModel
* variable. If ReplicationModel conflicts with distribution configuration, this
* function errors out.
* used depending on given distribution configuration.
*/
static char
DecideReplicationModel(char distributionMethod, bool viaDeprecatedAPI)
DecideReplicationModel(char distributionMethod, char *colocateWithTableName, bool
viaDeprecatedAPI)
{
if (viaDeprecatedAPI)
{
if (ReplicationModel != REPLICATION_MODEL_COORDINATOR)
{
ereport(NOTICE, (errmsg("using statement-based replication"),
errdetail("The current replication_model setting is "
"'streaming', which is not supported by "
"master_create_distributed_table."),
errhint("Use create_distributed_table to use the streaming "
"replication model.")));
}
return REPLICATION_MODEL_COORDINATOR;
}
else if (distributionMethod == DISTRIBUTE_BY_NONE)
{
return REPLICATION_MODEL_2PC;
}
else if (distributionMethod == DISTRIBUTE_BY_HASH)
else if (pg_strncasecmp(colocateWithTableName, "default", NAMEDATALEN) != 0 &&
!IsColocateWithNone(colocateWithTableName))
{
return ReplicationModel;
text *colocateWithTableNameText = cstring_to_text(colocateWithTableName);
Oid colocatedRelationId = ResolveRelationId(colocateWithTableNameText, false);
CitusTableCacheEntry *targetTableEntry = GetCitusTableCacheEntry(
colocatedRelationId);
char replicationModel = targetTableEntry->replicationModel;
return replicationModel;
}
else if (distributionMethod == DISTRIBUTE_BY_HASH &&
!DistributedTableReplicationIsEnabled())
{
return REPLICATION_MODEL_STREAMING;
}
else
{
if (ReplicationModel != REPLICATION_MODEL_COORDINATOR)
{
ereport(NOTICE, (errmsg("using statement-based replication"),
errdetail("Streaming replication is supported only for "
"hash-distributed tables.")));
}
return REPLICATION_MODEL_COORDINATOR;
}
@ -863,7 +851,6 @@ EnsureRelationCanBeDistributed(Oid relationId, Var *distributionColumn,
EnsureTableNotDistributed(relationId);
EnsureLocalTableEmptyIfNecessary(relationId, distributionMethod, viaDeprecatedAPI);
EnsureReplicationSettings(InvalidOid, replicationModel);
EnsureRelationHasNoTriggers(relationId);
/* we assume callers took necessary locks */
@ -1125,7 +1112,7 @@ static void
EnsureLocalTableEmpty(Oid relationId)
{
char *relationName = get_rel_name(relationId);
bool localTableEmpty = LocalTableEmpty(relationId);
bool localTableEmpty = TableEmpty(relationId);
if (!localTableEmpty)
{
@ -1156,36 +1143,6 @@ EnsureTableNotDistributed(Oid relationId)
}
/*
* EnsureReplicationSettings checks whether the current replication factor
* setting is compatible with the replication model. This function errors
* out if caller tries to use streaming replication with more than one
* replication factor.
*/
void
EnsureReplicationSettings(Oid relationId, char replicationModel)
{
char *msgSuffix = "the streaming replication model";
char *extraHint = " or setting \"citus.replication_model\" to \"statement\"";
if (relationId != InvalidOid)
{
msgSuffix = "tables which use the streaming replication model";
extraHint = "";
}
if (replicationModel == REPLICATION_MODEL_STREAMING &&
DistributedTableReplicationIsEnabled())
{
ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("replication factors above one are incompatible with %s",
msgSuffix),
errhint("Try again after reducing \"citus.shard_replication_"
"factor\" to one%s.", extraHint)));
}
}
/*
* EnsureRelationHasNoTriggers errors out if the given table has triggers on
* it. See also GetExplicitTriggerIdList function's comment for the triggers this
@ -1291,50 +1248,40 @@ SupportFunctionForColumn(Var *partitionColumn, Oid accessMethodId,
/*
* LocalTableEmpty function checks whether given local table contains any row and
* returns false if there is any data. This function is only for local tables and
* should not be called for distributed tables.
* TableEmpty function checks whether given table contains any row and
* returns false if there is any data.
*/
static bool
LocalTableEmpty(Oid tableId)
bool
TableEmpty(Oid tableId)
{
Oid schemaId = get_rel_namespace(tableId);
char *schemaName = get_namespace_name(schemaId);
char *tableName = get_rel_name(tableId);
char *tableQualifiedName = quote_qualified_identifier(schemaName, tableName);
StringInfo selectExistQueryString = makeStringInfo();
StringInfo selectTrueQueryString = makeStringInfo();
bool columnNull = false;
bool readOnly = true;
int rowId = 0;
int attributeId = 1;
AssertArg(!IsCitusTable(tableId));
int spiConnectionResult = SPI_connect();
if (spiConnectionResult != SPI_OK_CONNECT)
{
ereport(ERROR, (errmsg("could not connect to SPI manager")));
}
appendStringInfo(selectExistQueryString, SELECT_EXIST_QUERY, tableQualifiedName);
appendStringInfo(selectTrueQueryString, SELECT_TRUE_QUERY, tableQualifiedName);
int spiQueryResult = SPI_execute(selectExistQueryString->data, readOnly, 0);
int spiQueryResult = SPI_execute(selectTrueQueryString->data, readOnly, 0);
if (spiQueryResult != SPI_OK_SELECT)
{
ereport(ERROR, (errmsg("execution was not successful \"%s\"",
selectExistQueryString->data)));
selectTrueQueryString->data)));
}
/* we expect that SELECT EXISTS query will return single value in a single row */
Assert(SPI_processed == 1);
/* we expect that SELECT TRUE query will return single value in a single row OR empty set */
Assert(SPI_processed == 1 || SPI_processed == 0);
HeapTuple tuple = SPI_tuptable->vals[rowId];
Datum hasDataDatum = SPI_getbinval(tuple, SPI_tuptable->tupdesc, attributeId,
&columnNull);
bool localTableEmpty = !DatumGetBool(hasDataDatum);
bool localTableEmpty = !SPI_processed;
SPI_finish();

215
src/backend/distributed/commands/database.c Normal file
View File

@ -0,0 +1,215 @@
/*-------------------------------------------------------------------------
*
* database.c
* Commands to interact with the database object in a distributed
* environment.
*
* Copyright (c) Citus Data, Inc.
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#include "access/htup_details.h"
#include "access/xact.h"
#include "catalog/objectaddress.h"
#include "catalog/pg_database.h"
#include "commands/dbcommands.h"
#include "miscadmin.h"
#include "nodes/parsenodes.h"
#include "utils/syscache.h"
#include "distributed/commands.h"
#include "distributed/commands/utility_hook.h"
#include "distributed/deparser.h"
#include "distributed/metadata_sync.h"
#include "distributed/metadata_utility.h"
#include "distributed/multi_executor.h"
#include "distributed/relation_access_tracking.h"
#include "distributed/worker_transaction.h"
static void EnsureSequentialModeForDatabaseDDL(void);
static AlterOwnerStmt * RecreateAlterDatabaseOwnerStmt(Oid databaseOid);
static Oid get_database_owner(Oid db_oid);
/* controlled via GUC */
bool EnableAlterDatabaseOwner = false;
/*
* PreprocessAlterDatabaseOwnerStmt is called during the utility hook before the alter
* command is applied locally on the coordinator. This will verify if the command needs to
* be propagated to the workers and if so prepares a list of ddl commands to execute.
*/
List *
PreprocessAlterDatabaseOwnerStmt(Node *node, const char *queryString,
ProcessUtilityContext processUtilityContext)
{
AlterOwnerStmt *stmt = castNode(AlterOwnerStmt, node);
Assert(stmt->objectType == OBJECT_DATABASE);
ObjectAddress typeAddress = GetObjectAddressFromParseTree((Node *) stmt, false);
if (!ShouldPropagateObject(&typeAddress))
{
return NIL;
}
if (!EnableAlterDatabaseOwner)
{
/* don't propagate if GUC is turned off */
return NIL;
}
EnsureCoordinator();
QualifyTreeNode((Node *) stmt);
const char *sql = DeparseTreeNode((Node *) stmt);
EnsureSequentialModeForDatabaseDDL();
List *commands = list_make3(DISABLE_DDL_PROPAGATION,
(void *) sql,
ENABLE_DDL_PROPAGATION);
return NodeDDLTaskList(NON_COORDINATOR_NODES, commands);
}
/*
* PostprocessAlterDatabaseOwnerStmt is called during the utility hook after the alter
* database command has been applied locally.
*
* Its main purpose is to propagate the newly formed dependencies onto the nodes before
* applying the change of owner of the databse. This ensures, for systems that have role
* management, that the roles will be created before applying the alter owner command.
*/
List *
PostprocessAlterDatabaseOwnerStmt(Node *node, const char *queryString)
{
AlterOwnerStmt *stmt = castNode(AlterOwnerStmt, node);
Assert(stmt->objectType == OBJECT_DATABASE);
ObjectAddress typeAddress = GetObjectAddressFromParseTree((Node *) stmt, false);
if (!ShouldPropagateObject(&typeAddress))
{
return NIL;
}
if (!EnableAlterDatabaseOwner)
{
/* don't propagate if GUC is turned off */
return NIL;
}
EnsureDependenciesExistOnAllNodes(&typeAddress);
return NIL;
}
/*
* AlterDatabaseOwnerObjectAddress returns the ObjectAddress of the database that is the
* object of the AlterOwnerStmt. Errors if missing_ok is false.
*/
ObjectAddress
AlterDatabaseOwnerObjectAddress(Node *node, bool missing_ok)
{
AlterOwnerStmt *stmt = castNode(AlterOwnerStmt, node);
Assert(stmt->objectType == OBJECT_DATABASE);
Oid databaseOid = get_database_oid(strVal((Value *) stmt->object), missing_ok);
ObjectAddress address = { 0 };
ObjectAddressSet(address, DatabaseRelationId, databaseOid);
return address;
}
/*
* DatabaseOwnerDDLCommands returns a list of sql statements to idempotently apply a
* change of the database owner on the workers so that the database is owned by the same
* user on all nodes in the cluster.
*/
List *
DatabaseOwnerDDLCommands(const ObjectAddress *address)
{
Node *stmt = (Node *) RecreateAlterDatabaseOwnerStmt(address->objectId);
return list_make1(DeparseTreeNode(stmt));
}
/*
* RecreateAlterDatabaseOwnerStmt creates an AlterOwnerStmt that represents the operation
* of changing the owner of the database to its current owner.
*/
static AlterOwnerStmt *
RecreateAlterDatabaseOwnerStmt(Oid databaseOid)
{
AlterOwnerStmt *stmt = makeNode(AlterOwnerStmt);
stmt->objectType = OBJECT_DATABASE;
stmt->object = (Node *) makeString(get_database_name(databaseOid));
Oid ownerOid = get_database_owner(databaseOid);
stmt->newowner = makeNode(RoleSpec);
stmt->newowner->roletype = ROLESPEC_CSTRING;
stmt->newowner->rolename = GetUserNameFromId(ownerOid, false);
return stmt;
}
/*
* get_database_owner returns the Oid of the role owning the database
*/
static Oid
get_database_owner(Oid db_oid)
{
HeapTuple tuple = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(db_oid));
if (!HeapTupleIsValid(tuple))
{
ereport(ERROR, (errcode(ERRCODE_UNDEFINED_DATABASE),
errmsg("database with OID %u does not exist", db_oid)));
}
Oid dba = ((Form_pg_database) GETSTRUCT(tuple))->datdba;
ReleaseSysCache(tuple);
return dba;
}
/*
* EnsureSequentialModeForDatabaseDDL makes sure that the current transaction is already
* in sequential mode, or can still safely be put in sequential mode, it errors if that is
* not possible. The error contains information for the user to retry the transaction with
* sequential mode set from the beginning.
*/
static void
EnsureSequentialModeForDatabaseDDL(void)
{
if (!IsTransactionBlock())
{
/* we do not need to switch to sequential mode if we are not in a transaction */
return;
}
if (ParallelQueryExecutedInTransaction())
{
ereport(ERROR, (errmsg("cannot create or modify database because there was a "
"parallel operation on a distributed table in the "
"transaction"),
errdetail("When creating or altering a database, Citus needs to "
"perform all operations over a single connection per "
"node to ensure consistency."),
errhint("Try re-running the transaction with "
"\"SET LOCAL citus.multi_shard_modify_mode TO "
"\'sequential\';\"")));
}
ereport(DEBUG1, (errmsg("switching to sequential query execution mode"),
errdetail("Database is created or altered. To make sure subsequent "
"commands see the type correctly we need to make sure to "
"use only one connection for all future commands")));
SetLocalMultiShardModifyModeToSequential();
}

15
src/backend/distributed/commands/dependencies.c
View File

@ -14,6 +14,7 @@
#include "catalog/objectaddress.h"
#include "commands/extension.h"
#include "distributed/commands.h"
#include "distributed/commands/utility_hook.h"
#include "distributed/connection_management.h"
#include "distributed/listutils.h"
#include "distributed/metadata/dependency.h"
@ -191,6 +192,20 @@ GetDependencyCreateDDLCommands(const ObjectAddress *dependency)
return CreateCollationDDLsIdempotent(dependency->objectId);
}
case OCLASS_DATABASE:
{
List *databaseDDLCommands = NIL;
/* only propagate the ownership of the database when the feature is on */
if (EnableAlterDatabaseOwner)
{
List *ownerDDLCommands = DatabaseOwnerDDLCommands(dependency);
databaseDDLCommands = list_concat(databaseDDLCommands, ownerDDLCommands);
}
return databaseDDLCommands;
}
case OCLASS_PROC:
{
return CreateFunctionDDLCommandsIdempotent(dependency);

62
src/backend/distributed/commands/distribute_object_ops.c
View File

@ -240,6 +240,13 @@ static DistributeObjectOps Collation_Rename = {
.postprocess = NULL,
.address = RenameCollationStmtObjectAddress,
};
static DistributeObjectOps Database_AlterOwner = {
.deparse = DeparseAlterDatabaseOwnerStmt,
.qualify = NULL,
.preprocess = PreprocessAlterDatabaseOwnerStmt,
.postprocess = PostprocessAlterDatabaseOwnerStmt,
.address = AlterDatabaseOwnerObjectAddress,
};
static DistributeObjectOps Extension_AlterObjectSchema = {
.deparse = DeparseAlterExtensionSchemaStmt,
.qualify = NULL,
@ -359,6 +366,34 @@ static DistributeObjectOps Routine_AlterObjectDepends = {
.postprocess = NULL,
.address = AlterFunctionDependsStmtObjectAddress,
};
static DistributeObjectOps Sequence_Alter = {
.deparse = NULL,
.qualify = NULL,
.preprocess = PreprocessAlterSequenceStmt,
.postprocess = NULL,
.address = AlterSequenceObjectAddress,
};
static DistributeObjectOps Sequence_AlterObjectSchema = {
.deparse = NULL,
.qualify = NULL,
.preprocess = PreprocessAlterSequenceSchemaStmt,
.postprocess = NULL,
.address = AlterSequenceSchemaStmtObjectAddress,
};
static DistributeObjectOps Sequence_Drop = {
.deparse = DeparseDropSequenceStmt,
.qualify = NULL,
.preprocess = PreprocessDropSequenceStmt,
.postprocess = NULL,
.address = NULL,
};
static DistributeObjectOps Sequence_Rename = {
.deparse = DeparseRenameSequenceStmt,
.qualify = QualifyRenameSequenceStmt,
.preprocess = PreprocessRenameSequenceStmt,
.postprocess = NULL,
.address = RenameSequenceStmtObjectAddress,
};
static DistributeObjectOps Trigger_AlterObjectDepends = {
.deparse = NULL,
.qualify = NULL,
@ -453,7 +488,7 @@ static DistributeObjectOps Statistics_Rename = {
.address = NULL,
};
static DistributeObjectOps Table_AlterTable = {
.deparse = NULL,
.deparse = DeparseAlterTableStmt,
.qualify = NULL,
.preprocess = PreprocessAlterTableStmt,
.postprocess = NULL,
@ -621,6 +656,11 @@ GetDistributeObjectOps(Node *node)
return &Routine_AlterObjectSchema;
}
case OBJECT_SEQUENCE:
{
return &Sequence_AlterObjectSchema;
}
case OBJECT_STATISTIC_EXT:
{
return &Statistics_AlterObjectSchema;
@ -658,6 +698,11 @@ GetDistributeObjectOps(Node *node)
return &Collation_AlterOwner;
}
case OBJECT_DATABASE:
{
return &Database_AlterOwner;
}
case OBJECT_FUNCTION:
{
return &Function_AlterOwner;
@ -705,6 +750,11 @@ GetDistributeObjectOps(Node *node)
return &Any_AlterRoleSet;
}
case T_AlterSeqStmt:
{
return &Sequence_Alter;
}
#if PG_VERSION_NUM >= PG_VERSION_13
case T_AlterStatsStmt:
{
@ -861,6 +911,11 @@ GetDistributeObjectOps(Node *node)
return &Schema_Drop;
}
case OBJECT_SEQUENCE:
{
return &Sequence_Drop;
}
case OBJECT_STATISTIC_EXT:
{
return &Statistics_Drop;
@ -955,6 +1010,11 @@ GetDistributeObjectOps(Node *node)
return &Schema_Rename;
}
case OBJECT_SEQUENCE:
{
return &Sequence_Rename;
}
case OBJECT_STATISTIC_EXT:
{
return &Statistics_Rename;

8
src/backend/distributed/commands/drop_distributed_table.c
View File

@ -59,6 +59,8 @@ master_drop_distributed_table_metadata(PG_FUNCTION_ARGS)
Datum
master_remove_partition_metadata(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *schemaNameText = PG_GETARG_TEXT_P(1);
text *tableNameText = PG_GETARG_TEXT_P(2);
@ -66,8 +68,6 @@ master_remove_partition_metadata(PG_FUNCTION_ARGS)
char *schemaName = text_to_cstring(schemaNameText);
char *tableName = text_to_cstring(tableNameText);
CheckCitusVersion(ERROR);
/*
* The SQL_DROP trigger calls this function even for tables that are
* not distributed. In that case, silently ignore. This is not very
@ -97,6 +97,8 @@ master_remove_partition_metadata(PG_FUNCTION_ARGS)
Datum
master_remove_distributed_table_metadata_from_workers(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *schemaNameText = PG_GETARG_TEXT_P(1);
text *tableNameText = PG_GETARG_TEXT_P(2);
@ -104,8 +106,6 @@ master_remove_distributed_table_metadata_from_workers(PG_FUNCTION_ARGS)
char *schemaName = text_to_cstring(schemaNameText);
char *tableName = text_to_cstring(tableNameText);
CheckCitusVersion(ERROR);
CheckTableSchemaNameForDrop(relationId, &schemaName, &tableName);
MasterRemoveDistributedTableMetadataFromWorkers(relationId, schemaName, tableName);

43
src/backend/distributed/commands/foreign_constraint.c
View File

@ -734,8 +734,8 @@ HasForeignKeyWithLocalTable(Oid relationId)
/*
* GetForeignKeysWithLocalTables returns a list foreign keys for foreign key
* relationaships that relation has with local tables.
* GetForeignKeysWithLocalTables returns a list of foreign keys for foreign key
* relationships that relation has with local tables.
*/
static List *
GetForeignKeysWithLocalTables(Oid relationId)
@ -753,6 +753,21 @@ GetForeignKeysWithLocalTables(Oid relationId)
}
/*
* GetForeignKeysFromLocalTables returns a list of foreign keys where the referencing
* relation is a local table.
*/
List *
GetForeignKeysFromLocalTables(Oid relationId)
{
int referencedFKeysFlag = INCLUDE_REFERENCED_CONSTRAINTS |
INCLUDE_LOCAL_TABLES;
List *referencingFKeyList = GetForeignKeyOids(relationId, referencedFKeysFlag);
return referencingFKeyList;
}
/*
* HasForeignKeyToCitusLocalTable returns true if any of the foreign key constraints
* on the relation with relationId references to a citus local table.
@ -1102,6 +1117,30 @@ GetReferencedTableId(Oid foreignKeyId)
}
/*
* GetReferencingTableId returns OID of the referencing relation for the foreign
* key with foreignKeyId. If there is no such foreign key, then this function
* returns InvalidOid.
*/
Oid
GetReferencingTableId(Oid foreignKeyId)
{
HeapTuple heapTuple = SearchSysCache1(CONSTROID, ObjectIdGetDatum(foreignKeyId));
if (!HeapTupleIsValid(heapTuple))
{
/* no such foreign key */
return InvalidOid;
}
Form_pg_constraint constraintForm = (Form_pg_constraint) GETSTRUCT(heapTuple);
Oid referencingTableId = constraintForm->conrelid;
ReleaseSysCache(heapTuple);
return referencingTableId;
}
/*
* IsTableTypeIncluded returns true if type of the table with relationId (distributed,
* reference, Citus local or Postgres local) is included in the flags, false if not

13
src/backend/distributed/commands/function.c
View File

@ -461,15 +461,6 @@ GetFunctionColocationId(Oid functionOid, char *colocateWithTableName,
EnsureFunctionCanBeColocatedWithTable(functionOid, distributionArgumentOid,
colocatedTableId);
}
else if (ReplicationModel == REPLICATION_MODEL_COORDINATOR)
{
/* streaming replication model is required for metadata syncing */
ereport(ERROR, (errmsg("cannot create a function with a distribution "
"argument when citus.replication_model is "
"'statement'"),
errhint("Set citus.replication_model to 'streaming' "
"before creating distributed tables")));
}
}
else
{
@ -537,7 +528,7 @@ EnsureFunctionCanBeColocatedWithTable(Oid functionOid, Oid distributionColumnTyp
"with distributed tables that are created using "
"streaming replication model."),
errhint("When distributing tables make sure that "
"citus.replication_model = 'streaming'")));
"citus.shard_replication_factor = 1")));
}
/*
@ -1438,7 +1429,7 @@ PreprocessAlterFunctionSchemaStmt(Node *node, const char *queryString,
/*
* PreprocessAlterTypeOwnerStmt is called for change of owner ship of functions before the owner
* PreprocessAlterFunctionOwnerStmt is called for change of owner ship of functions before the owner
* ship is changed on the local instance.
*
* If the function for which the owner is changed is distributed we execute the change on

5
src/backend/distributed/commands/index.c
View File

@ -302,7 +302,8 @@ CreateIndexStmtGetSchemaId(IndexStmt *createIndexStatement)
* It returns a list that is filled by the pgIndexProcessor.
*/
List *
ExecuteFunctionOnEachTableIndex(Oid relationId, PGIndexProcessor pgIndexProcessor)
ExecuteFunctionOnEachTableIndex(Oid relationId, PGIndexProcessor pgIndexProcessor,
int indexFlags)
{
List *result = NIL;
ScanKeyData scanKey[1];
@ -324,7 +325,7 @@ ExecuteFunctionOnEachTableIndex(Oid relationId, PGIndexProcessor pgIndexProcesso
while (HeapTupleIsValid(heapTuple))
{
Form_pg_index indexForm = (Form_pg_index) GETSTRUCT(heapTuple);
pgIndexProcessor(indexForm, &result);
pgIndexProcessor(indexForm, &result, indexFlags);
heapTuple = systable_getnext(scanDescriptor);
}

2
src/backend/distributed/commands/multi_copy.c
View File

@ -2228,7 +2228,7 @@ CitusCopyDestReceiverStartup(DestReceiver *dest, int operation,
if (cacheEntry->replicationModel == REPLICATION_MODEL_2PC ||
MultiShardCommitProtocol == COMMIT_PROTOCOL_2PC)
{
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
}
/* define how tuples will be serialised */

317
src/backend/distributed/commands/sequence.c
View File

@ -14,14 +14,20 @@
#include "catalog/dependency.h"
#include "catalog/namespace.h"
#include "commands/defrem.h"
#include "commands/extension.h"
#include "distributed/commands.h"
#include "distributed/commands/sequence.h"
#include "distributed/commands/utility_hook.h"
#include "distributed/deparser.h"
#include "distributed/listutils.h"
#include "distributed/metadata/distobject.h"
#include "distributed/metadata_cache.h"
#include "distributed/metadata_sync.h"
#include "nodes/parsenodes.h"
/* Local functions forward declarations for helper functions */
static bool OptionsSpecifyOwnedBy(List *optionList, Oid *ownedByTableId);
static bool ShouldPropagateAlterSequence(const ObjectAddress *address);
/*
@ -92,15 +98,6 @@ ErrorIfDistributedAlterSeqOwnedBy(AlterSeqStmt *alterSeqStmt)
errmsg("cannot alter OWNED BY option of a sequence "
"already owned by a distributed table")));
}
else if (!hasDistributedOwner && IsCitusTable(newOwnedByTableId))
{
/* and don't let local sequences get a distributed OWNED BY */
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot associate an existing sequence with a "
"distributed table"),
errhint("Use a sequence in a distributed table by specifying "
"a serial column type before creating any shards.")));
}
}
}
@ -209,3 +206,305 @@ ExtractDefaultColumnsAndOwnedSequences(Oid relationId, List **columnNameList,
relation_close(relation, NoLock);
}
/*
* PreprocessDropSequenceStmt gets called during the planning phase of a DROP SEQUENCE statement
* and returns a list of DDLJob's that will drop any distributed sequences from the
* workers.
*
* The DropStmt could have multiple objects to drop, the list of objects will be filtered
* to only keep the distributed sequences for deletion on the workers. Non-distributed
* sequences will still be dropped locally but not on the workers.
*/
List *
PreprocessDropSequenceStmt(Node *node, const char *queryString,
ProcessUtilityContext processUtilityContext)
{
DropStmt *stmt = castNode(DropStmt, node);
List *deletingSequencesList = stmt->objects;
List *distributedSequencesList = NIL;
List *distributedSequenceAddresses = NIL;
Assert(stmt->removeType == OBJECT_SEQUENCE);
if (creating_extension)
{
/*
* extensions should be created separately on the workers, sequences cascading
* from an extension should therefor not be propagated here.
*/
return NIL;
}
if (!EnableDependencyCreation)
{
/*
* we are configured to disable object propagation, should not propagate anything
*/
return NIL;
}
/*
* Our statements need to be fully qualified so we can drop them from the right schema
* on the workers
*/
QualifyTreeNode((Node *) stmt);
/*
* iterate over all sequences to be dropped and filter to keep only distributed
* sequences.
*/
List *objectNameList = NULL;
foreach_ptr(objectNameList, deletingSequencesList)
{
RangeVar *seq = makeRangeVarFromNameList(objectNameList);
Oid seqOid = RangeVarGetRelid(seq, NoLock, stmt->missing_ok);
ObjectAddress sequenceAddress = { 0 };
ObjectAddressSet(sequenceAddress, RelationRelationId, seqOid);
if (!IsObjectDistributed(&sequenceAddress))
{
continue;
}
/* collect information for all distributed sequences */
ObjectAddress *addressp = palloc(sizeof(ObjectAddress));
*addressp = sequenceAddress;
distributedSequenceAddresses = lappend(distributedSequenceAddresses, addressp);
distributedSequencesList = lappend(distributedSequencesList, objectNameList);
}
if (list_length(distributedSequencesList) <= 0)
{
/* no distributed functions to drop */
return NIL;
}
/*
* managing types can only be done on the coordinator if ddl propagation is on. when
* it is off we will never get here. MX workers don't have a notion of distributed
* types, so we block the call.
*/
EnsureCoordinator();
/* remove the entries for the distributed objects on dropping */
ObjectAddress *address = NULL;
foreach_ptr(address, distributedSequenceAddresses)
{
UnmarkObjectDistributed(address);
}
/*
* Swap the list of objects before deparsing and restore the old list after. This
* ensures we only have distributed sequences in the deparsed drop statement.
*/
DropStmt *stmtCopy = copyObject(stmt);
stmtCopy->objects = distributedSequencesList;
stmtCopy->missing_ok = true;
const char *dropStmtSql = DeparseTreeNode((Node *) stmtCopy);
List *commands = list_make3(DISABLE_DDL_PROPAGATION,
(void *) dropStmtSql,
ENABLE_DDL_PROPAGATION);
return NodeDDLTaskList(NON_COORDINATOR_NODES, commands);
}
/*
* PreprocessRenameSequenceStmt is called when the user is renaming a sequence. The invocation
* happens before the statement is applied locally.
*
* As the sequence already exists we have access to the ObjectAddress, this is used to
* check if it is distributed. If so the rename is executed on all the workers to keep the
* types in sync across the cluster.
*/
List *
PreprocessRenameSequenceStmt(Node *node, const char *queryString, ProcessUtilityContext
processUtilityContext)
{
RenameStmt *stmt = castNode(RenameStmt, node);
Assert(stmt->renameType == OBJECT_SEQUENCE);
ObjectAddress address = GetObjectAddressFromParseTree((Node *) stmt,
stmt->missing_ok);
if (!ShouldPropagateAlterSequence(&address))
{
return NIL;
}
EnsureCoordinator();
QualifyTreeNode((Node *) stmt);
/* this takes care of cases where not all workers have synced metadata */
RenameStmt *stmtCopy = copyObject(stmt);
stmtCopy->missing_ok = true;
const char *sql = DeparseTreeNode((Node *) stmtCopy);
List *commands = list_make3(DISABLE_DDL_PROPAGATION, (void *) sql,
ENABLE_DDL_PROPAGATION);
return NodeDDLTaskList(NON_COORDINATOR_NODES, commands);
}
/*
* RenameSequenceStmtObjectAddress returns the ObjectAddress of the sequence that is the
* subject of the RenameStmt.
*/
ObjectAddress
RenameSequenceStmtObjectAddress(Node *node, bool missing_ok)
{
RenameStmt *stmt = castNode(RenameStmt, node);
Assert(stmt->renameType == OBJECT_SEQUENCE);
RangeVar *sequence = stmt->relation;
Oid seqOid = RangeVarGetRelid(sequence, NoLock, missing_ok);
ObjectAddress sequenceAddress = { 0 };
ObjectAddressSet(sequenceAddress, RelationRelationId, seqOid);
return sequenceAddress;
}
/*
* ShouldPropagateAlterSequence returns, based on the address of a sequence, if alter
* statements targeting the function should be propagated.
*/
static bool
ShouldPropagateAlterSequence(const ObjectAddress *address)
{
if (creating_extension)
{
/*
* extensions should be created separately on the workers, sequences cascading
* from an extension should therefore not be propagated.
*/
return false;
}
if (!EnableDependencyCreation)
{
/*
* we are configured to disable object propagation, should not propagate anything
*/
return false;
}
if (!IsObjectDistributed(address))
{
/* do not propagate alter sequence for non-distributed sequences */
return false;
}
return true;
}
/*
* PreprocessAlterSequenceStmt gets called during the planning phase of an ALTER SEQUENCE statement
* of one of the following forms:
* ALTER SEQUENCE [ IF EXISTS ] name
* [ AS data_type ]
* [ INCREMENT [ BY ] increment ]
* [ MINVALUE minvalue | NO MINVALUE ] [ MAXVALUE maxvalue | NO MAXVALUE ]
* [ START [ WITH ] start ]
* [ RESTART [ [ WITH ] restart ] ]
* [ CACHE cache ] [ [ NO ] CYCLE ]
* [ OWNED BY { table_name.column_name | NONE } ]
*
* For distributed sequences, this operation will not be allowed for now.
* The reason is that we change sequence parameters when distributing it, so we don't want to
* touch those parameters for now.
*/
List *
PreprocessAlterSequenceStmt(Node *node, const char *queryString,
ProcessUtilityContext processUtilityContext)
{
AlterSeqStmt *stmt = castNode(AlterSeqStmt, node);
ObjectAddress address = GetObjectAddressFromParseTree((Node *) stmt,
stmt->missing_ok);
/* error out if the sequence is distributed */
if (IsObjectDistributed(&address))
{
ereport(ERROR, (errmsg(
"This operation is currently not allowed for a distributed sequence.")));
}
else
{
return NIL;
}
}
/*
* AlterSequenceOwnerObjectAddress returns the ObjectAddress of the sequence that is the
* subject of the AlterOwnerStmt.
*/
ObjectAddress
AlterSequenceObjectAddress(Node *node, bool missing_ok)
{
AlterSeqStmt *stmt = castNode(AlterSeqStmt, node);
RangeVar *sequence = stmt->sequence;
Oid seqOid = RangeVarGetRelid(sequence, NoLock, stmt->missing_ok);
ObjectAddress sequenceAddress = { 0 };
ObjectAddressSet(sequenceAddress, RelationRelationId, seqOid);
return sequenceAddress;
}
/*
* PreprocessAlterSequenceSchemaStmt is executed before the statement is applied to the local
* postgres instance.
*
* For distributed sequences, this operation will not be allowed for now.
*/
List *
PreprocessAlterSequenceSchemaStmt(Node *node, const char *queryString,
ProcessUtilityContext processUtilityContext)
{
AlterObjectSchemaStmt *stmt = castNode(AlterObjectSchemaStmt, node);
Assert(stmt->objectType == OBJECT_SEQUENCE);
ObjectAddress address = GetObjectAddressFromParseTree((Node *) stmt,
stmt->missing_ok);
/* error out if the sequence is distributed */
if (IsObjectDistributed(&address))
{
ereport(ERROR, (errmsg(
"This operation is currently not allowed for a distributed sequence.")));
}
else
{
return NIL;
}
}
/*
* AlterSequenceSchemaStmtObjectAddress returns the ObjectAddress of the sequence that is
* the subject of the AlterObjectSchemaStmt.
*/
ObjectAddress
AlterSequenceSchemaStmtObjectAddress(Node *node, bool missing_ok)
{
AlterObjectSchemaStmt *stmt = castNode(AlterObjectSchemaStmt, node);
Assert(stmt->objectType == OBJECT_SEQUENCE);
RangeVar *sequence = stmt->relation;
Oid seqOid = RangeVarGetRelid(sequence, NoLock, missing_ok);
ObjectAddress sequenceAddress = { 0 };
ObjectAddressSet(sequenceAddress, RelationRelationId, seqOid);
return sequenceAddress;
}

371
src/backend/distributed/commands/table.c
View File

@ -17,6 +17,7 @@
#include "catalog/pg_class.h"
#include "catalog/pg_constraint.h"
#include "catalog/pg_depend.h"
#include "catalog/pg_type.h"
#include "commands/tablecmds.h"
#include "distributed/citus_ruleutils.h"
#include "distributed/colocation_utils.h"
@ -28,14 +29,17 @@
#include "distributed/coordinator_protocol.h"
#include "distributed/metadata_sync.h"
#include "distributed/metadata/dependency.h"
#include "distributed/metadata/distobject.h"
#include "distributed/multi_executor.h"
#include "distributed/multi_partitioning_utils.h"
#include "distributed/reference_table_utils.h"
#include "distributed/relation_access_tracking.h"
#include "distributed/resource_lock.h"
#include "distributed/version_compat.h"
#include "distributed/worker_shard_visibility.h"
#include "lib/stringinfo.h"
#include "nodes/parsenodes.h"
#include "parser/parse_expr.h"
#include "storage/lmgr.h"
#include "utils/builtins.h"
#include "utils/lsyscache.h"
@ -122,6 +126,8 @@ PreprocessDropTableStmt(Node *node, const char *queryString,
Oid relationId = RangeVarGetRelid(tableRangeVar, AccessShareLock, missingOK);
ErrorIfIllegallyChangingKnownShard(relationId);
/* we're not interested in non-valid, non-distributed relations */
if (relationId == InvalidOid || !IsCitusTable(relationId))
{
@ -165,6 +171,8 @@ PreprocessDropTableStmt(Node *node, const char *queryString,
SendCommandToWorkersWithMetadata(detachPartitionCommand);
}
SendCommandToWorkersWithMetadata(ENABLE_DDL_PROPAGATION);
}
return NIL;
@ -574,24 +582,50 @@ PreprocessAlterTableStmt(Node *node, const char *alterTableCommand,
ErrorIfUnsupportedAlterTableStmt(alterTableStatement);
}
EnsureCoordinator();
/* these will be set in below loop according to subcommands */
Oid rightRelationId = InvalidOid;
bool executeSequentially = false;
/*
* We check if there is a ADD/DROP FOREIGN CONSTRAINT command in sub commands
* We check if there is:
* - an ADD/DROP FOREIGN CONSTRAINT command in sub commands
* list. If there is we assign referenced relation id to rightRelationId and
* we also set skip_validation to true to prevent PostgreSQL to verify validity
* of the foreign constraint in master. Validity will be checked in workers
* anyway.
* - an ADD COLUMN .. DEFAULT nextval('..') OR
* an ADD COLUMN .. SERIAL pseudo-type OR
* an ALTER COLUMN .. SET DEFAULT nextval('..'). If there is we set
* deparseAT variable to true which means we will deparse the statement
* before we propagate the command to shards. For shards, all the defaults
* coming from a user-defined sequence will be replaced by
* NOT NULL constraint.
*/
List *commandList = alterTableStatement->cmds;
/*
* if deparsing is needed, we will use a different version of the original
* alterTableStmt
*/
bool deparseAT = false;
bool propagateCommandToWorkers = true;
AlterTableStmt *newStmt = copyObject(alterTableStatement);
AlterTableCmd *newCmd = makeNode(AlterTableCmd);
AlterTableCmd *command = NULL;
foreach_ptr(command, commandList)
{
AlterTableType alterTableType = command->subtype;
/*
* if deparsing is needed, we will use a different version of the original
* AlterTableCmd
*/
newCmd = copyObject(command);
if (alterTableType == AT_AddConstraint)
{
Constraint *constraint = (Constraint *) command->def;
@ -666,6 +700,96 @@ PreprocessAlterTableStmt(Node *node, const char *alterTableCommand,
break;
}
}
/*
* We check for ADD COLUMN .. DEFAULT expr
* if expr contains nextval('user_defined_seq')
* we should deparse the statement
*/
constraint = NULL;
foreach_ptr(constraint, columnConstraints)
{
if (constraint->contype == CONSTR_DEFAULT)
{
if (constraint->raw_expr != NULL)
{
ParseState *pstate = make_parsestate(NULL);
Node *expr = transformExpr(pstate, constraint->raw_expr,
EXPR_KIND_COLUMN_DEFAULT);
if (contain_nextval_expression_walker(expr, NULL))
{
deparseAT = true;
/* the new column definition will have no constraint */
ColumnDef *newColDef = copyObject(columnDefinition);
newColDef->constraints = NULL;
newCmd->def = (Node *) newColDef;
}
}
}
}
/*
* We check for ADD COLUMN .. SERIAL pseudo-type
* if that's the case, we should deparse the statement
* The structure of this check is copied from transformColumnDefinition.
*/
if (columnDefinition->typeName && list_length(
columnDefinition->typeName->names) == 1 &&
!columnDefinition->typeName->pct_type)
{
char *typeName = strVal(linitial(columnDefinition->typeName->names));
if (strcmp(typeName, "smallserial") == 0 ||
strcmp(typeName, "serial2") == 0 ||
strcmp(typeName, "serial") == 0 ||
strcmp(typeName, "serial4") == 0 ||
strcmp(typeName, "bigserial") == 0 ||
strcmp(typeName, "serial8") == 0)
{
deparseAT = true;
ColumnDef *newColDef = copyObject(columnDefinition);
newColDef->is_not_null = false;
if (strcmp(typeName, "smallserial") == 0 ||
strcmp(typeName, "serial2") == 0)
{
newColDef->typeName->names = NIL;
newColDef->typeName->typeOid = INT2OID;
}
else if (strcmp(typeName, "serial") == 0 ||
strcmp(typeName, "serial4") == 0)
{
newColDef->typeName->names = NIL;
newColDef->typeName->typeOid = INT4OID;
}
else if (strcmp(typeName, "bigserial") == 0 ||
strcmp(typeName, "serial8") == 0)
{
newColDef->typeName->names = NIL;
newColDef->typeName->typeOid = INT8OID;
}
newCmd->def = (Node *) newColDef;
}
}
}
/*
* We check for ALTER COLUMN .. SET/DROP DEFAULT
* we should not propagate anything to shards
*/
else if (alterTableType == AT_ColumnDefault)
{
ParseState *pstate = make_parsestate(NULL);
Node *expr = transformExpr(pstate, command->def,
EXPR_KIND_COLUMN_DEFAULT);
if (contain_nextval_expression_walker(expr, NULL))
{
propagateCommandToWorkers = false;
}
}
else if (alterTableType == AT_AttachPartition)
{
@ -731,12 +855,20 @@ PreprocessAlterTableStmt(Node *node, const char *alterTableCommand,
DDLJob *ddlJob = palloc0(sizeof(DDLJob));
ddlJob->targetRelationId = leftRelationId;
ddlJob->concurrentIndexCmd = false;
const char *sqlForTaskList = alterTableCommand;
if (deparseAT)
{
newStmt->cmds = list_make1(newCmd);
sqlForTaskList = DeparseTreeNode((Node *) newStmt);
}
ddlJob->commandString = alterTableCommand;
if (OidIsValid(rightRelationId))
{
bool referencedIsLocalTable = !IsCitusTable(rightRelationId);
if (referencedIsLocalTable)
if (referencedIsLocalTable || !propagateCommandToWorkers)
{
ddlJob->taskList = NIL;
}
@ -744,13 +876,17 @@ PreprocessAlterTableStmt(Node *node, const char *alterTableCommand,
{
/* if foreign key related, use specialized task list function ... */
ddlJob->taskList = InterShardDDLTaskList(leftRelationId, rightRelationId,
alterTableCommand);
sqlForTaskList);
}
}
else
{
/* ... otherwise use standard DDL task list function */
ddlJob->taskList = DDLTaskList(leftRelationId, alterTableCommand);
ddlJob->taskList = DDLTaskList(leftRelationId, sqlForTaskList);
if (!propagateCommandToWorkers)
{
ddlJob->taskList = NIL;
}
}
List *ddlJobs = list_make1(ddlJob);
@ -1467,8 +1603,80 @@ PostprocessAlterTableStmt(AlterTableStmt *alterTableStatement)
constraint);
}
}
/*
* We check for ADD COLUMN .. DEFAULT expr
* if expr contains nextval('user_defined_seq')
* we should make sure that the type of the column that uses
* that sequence is supported
*/
constraint = NULL;
foreach_ptr(constraint, columnConstraints)
{
if (constraint->contype == CONSTR_DEFAULT)
{
if (constraint->raw_expr != NULL)
{
ParseState *pstate = make_parsestate(NULL);
Node *expr = transformExpr(pstate, constraint->raw_expr,
EXPR_KIND_COLUMN_DEFAULT);
/*
* We should make sure that the type of the column that uses
* that sequence is supported
*/
if (contain_nextval_expression_walker(expr, NULL))
{
AttrNumber attnum = get_attnum(relationId,
columnDefinition->colname);
Oid seqTypId = GetAttributeTypeOid(relationId, attnum);
EnsureSequenceTypeSupported(relationId, attnum, seqTypId);
}
}
}
}
}
/*
* We check for ALTER COLUMN .. SET DEFAULT nextval('user_defined_seq')
* we should make sure that the type of the column that uses
* that sequence is supported
*/
else if (alterTableType == AT_ColumnDefault)
{
ParseState *pstate = make_parsestate(NULL);
Node *expr = transformExpr(pstate, command->def,
EXPR_KIND_COLUMN_DEFAULT);
if (contain_nextval_expression_walker(expr, NULL))
{
AttrNumber attnum = get_attnum(relationId, command->name);
Oid seqTypId = GetAttributeTypeOid(relationId, attnum);
EnsureSequenceTypeSupported(relationId, attnum, seqTypId);
}
}
}
/* for the new sequences coming with this ALTER TABLE statement */
if (ShouldSyncTableMetadata(relationId) && ClusterHasKnownMetadataWorkers())
{
List *sequenceCommandList = NIL;
/* commands to create sequences */
List *sequenceDDLCommands = SequenceDDLCommandsForTable(relationId);
sequenceCommandList = list_concat(sequenceCommandList, sequenceDDLCommands);
/* prevent recursive propagation */
SendCommandToWorkersWithMetadata(DISABLE_DDL_PROPAGATION);
/* send the commands one by one */
const char *sequenceCommand = NULL;
foreach_ptr(sequenceCommand, sequenceCommandList)
{
SendCommandToWorkersWithMetadata(sequenceCommand);
}
SendCommandToWorkersWithMetadata(ENABLE_DDL_PROPAGATION);
}
}
@ -1735,10 +1943,101 @@ ErrorIfUnsupportedAlterTableStmt(AlterTableStmt *alterTableStatement)
strcmp(typeName, "serial4") == 0 ||
strcmp(typeName, "bigserial") == 0 ||
strcmp(typeName, "serial8") == 0)
{
/*
* We currently don't support adding a serial column for an MX table
* TODO: record the dependency in the workers
*/
if (ShouldSyncTableMetadata(relationId) &&
ClusterHasKnownMetadataWorkers())
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot execute ADD COLUMN commands "
"involving serial pseudotypes")));
errmsg(
"cannot execute ADD COLUMN commands involving serial"
" pseudotypes when metadata is synchronized to workers")));
}
/*
* we only allow adding a serial column if it is the only subcommand
* and it has no constraints
*/
if (commandList->length > 1 || column->constraints)
{
ereport(ERROR, (errcode(
ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg(
"cannot execute ADD COLUMN commands involving "
"serial pseudotypes with other subcommands/constraints"),
errhint(
"You can issue each subcommand separately")));
}
/*
* Currently we don't support backfilling the new column with default values
* if the table is not empty
*/
if (!TableEmpty(relationId))
{
ereport(ERROR, (errcode(
ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg(
"Cannot add a column involving serial pseudotypes "
"because the table is not empty"),
errhint(
"You can first call ALTER TABLE .. ADD COLUMN .. smallint/int/bigint\n"
"Then set the default by ALTER TABLE .. ALTER COLUMN .. SET DEFAULT nextval('..')")));
}
}
}
List *columnConstraints = column->constraints;
Constraint *constraint = NULL;
foreach_ptr(constraint, columnConstraints)
{
if (constraint->contype == CONSTR_DEFAULT)
{
if (constraint->raw_expr != NULL)
{
ParseState *pstate = make_parsestate(NULL);
Node *expr = transformExpr(pstate, constraint->raw_expr,
EXPR_KIND_COLUMN_DEFAULT);
if (contain_nextval_expression_walker(expr, NULL))
{
/*
* we only allow adding a column with non_const default
* if its the only subcommand and has no other constraints
*/
if (commandList->length > 1 ||
columnConstraints->length > 1)
{
ereport(ERROR, (errcode(
ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg(
"cannot execute ADD COLUMN .. DEFAULT nextval('..')"
" command with other subcommands/constraints"),
errhint(
"You can issue each subcommand separately")));
}
/*
* Currently we don't support backfilling the new column with default values
* if the table is not empty
*/
if (!TableEmpty(relationId))
{
ereport(ERROR, (errcode(
ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg(
"cannot add a column involving DEFAULT nextval('..') "
"because the table is not empty"),
errhint(
"You can first call ALTER TABLE .. ADD COLUMN .. smallint/int/bigint\n"
"Then set the default by ALTER TABLE .. ALTER COLUMN .. SET DEFAULT nextval('..')")));
}
}
}
}
}
}
@ -1746,9 +2045,67 @@ ErrorIfUnsupportedAlterTableStmt(AlterTableStmt *alterTableStatement)
break;
}
case AT_DropColumn:
case AT_ColumnDefault:
{
if (AlterInvolvesPartitionColumn(alterTableStatement, command))
{
ereport(ERROR, (errmsg("cannot execute ALTER TABLE command "
"involving partition column")));
}
ParseState *pstate = make_parsestate(NULL);
Node *expr = transformExpr(pstate, command->def,
EXPR_KIND_COLUMN_DEFAULT);
if (contain_nextval_expression_walker(expr, NULL))
{
/*
* we only allow altering a column's default to non_const expr
* if its the only subcommand
*/
if (commandList->length > 1)
{
ereport(ERROR, (errcode(
ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg(
"cannot execute ALTER COLUMN COLUMN .. SET DEFAULT "
"nextval('..') command with other subcommands"),
errhint(
"You can issue each subcommand separately")));
}
}
break;
}
case AT_AlterColumnType:
{
if (AlterInvolvesPartitionColumn(alterTableStatement, command))
{
ereport(ERROR, (errmsg("cannot execute ALTER TABLE command "
"involving partition column")));
}
/*
* We check for ALTER COLUMN TYPE ...
* if the column has default coming from a user-defined sequence
* changing the type of the column should not be allowed for now
*/
AttrNumber attnum = get_attnum(relationId, command->name);
List *attnumList = NIL;
List *dependentSequenceList = NIL;
GetDependentSequencesWithRelation(relationId, &attnumList,
&dependentSequenceList, attnum);
if (dependentSequenceList != NIL)
{
ereport(ERROR, (errmsg("cannot execute ALTER COLUMN TYPE .. command "
"because the column involves a default coming "
"from a sequence")));
}
break;
}
case AT_DropColumn:
case AT_DropNotNull:
{
if (AlterInvolvesPartitionColumn(alterTableStatement, command))

24
src/backend/distributed/commands/truncate.c
View File

@ -32,6 +32,7 @@
#include "distributed/resource_lock.h"
#include "distributed/transaction_management.h"
#include "distributed/worker_transaction.h"
#include "distributed/worker_shard_visibility.h"
#include "storage/lmgr.h"
#include "utils/builtins.h"
#include "utils/lsyscache.h"
@ -174,10 +175,11 @@ TruncateTaskList(Oid relationId)
Datum
truncate_local_data_after_distributing_table(PG_FUNCTION_ARGS)
{
Oid relationId = PG_GETARG_OID(0);
CheckCitusVersion(ERROR);
EnsureCoordinator();
Oid relationId = PG_GETARG_OID(0);
EnsureLocalTableCanBeTruncated(relationId);
TruncateStmt *truncateStmt = makeNode(TruncateStmt);
@ -215,16 +217,12 @@ EnsureLocalTableCanBeTruncated(Oid relationId)
"tables.")));
}
/* make sure there are no foreign key references from a local table */
SetForeignConstraintRelationshipGraphInvalid();
List *referencingRelationList = ReferencingRelationIdList(relationId);
Oid referencingRelation = InvalidOid;
foreach_oid(referencingRelation, referencingRelationList)
{
/* we do not truncate a table if there is a local table referencing it */
if (!IsCitusTable(referencingRelation))
List *referencingForeignConstaintsFromLocalTables =
GetForeignKeysFromLocalTables(relationId);
if (list_length(referencingForeignConstaintsFromLocalTables) > 0)
{
Oid foreignKeyId = linitial_oid(referencingForeignConstaintsFromLocalTables);
Oid referencingRelation = GetReferencingTableId(foreignKeyId);
char *referencedRelationName = get_rel_name(relationId);
char *referencingRelationName = get_rel_name(referencingRelation);
@ -234,7 +232,6 @@ EnsureLocalTableCanBeTruncated(Oid relationId)
referencingRelationName,
referencedRelationName)));
}
}
}
@ -265,6 +262,9 @@ ErrorIfUnsupportedTruncateStmt(TruncateStmt *truncateStatement)
foreach_ptr(rangeVar, relationList)
{
Oid relationId = RangeVarGetRelid(rangeVar, NoLock, false);
ErrorIfIllegallyChangingKnownShard(relationId);
char relationKind = get_rel_relkind(relationId);
if (IsCitusTable(relationId) &&
relationKind == RELKIND_FOREIGN_TABLE)

1
src/backend/distributed/connection/connection_configuration.c
View File

@ -21,6 +21,7 @@
/* stores the string representation of our node connection GUC */
char *NodeConninfo = "";
char *LocalHostName = "localhost";
/* represents a list of libpq parameter settings */
typedef struct ConnParamsInfo

4
src/backend/distributed/connection/locally_reserved_shared_connections.c
View File

@ -111,9 +111,9 @@ PG_FUNCTION_INFO_V1(citus_reserved_connection_stats);
Datum
citus_reserved_connection_stats(PG_FUNCTION_ARGS)
{
TupleDesc tupleDescriptor = NULL;
CheckCitusVersion(ERROR);
TupleDesc tupleDescriptor = NULL;
Tuplestorestate *tupleStore = SetupTuplestore(fcinfo, &tupleDescriptor);
StoreAllReservedConnections(tupleStore, tupleDescriptor);

12
src/backend/distributed/connection/placement_connection.c
View File

@ -846,6 +846,18 @@ ConnectionModifiedPlacement(MultiConnection *connection)
return false;
}
if (dlist_is_empty(&connection->referencedPlacements))
{
/*
* When referencesPlacements are empty, it means that we come here
* from an API that uses a node connection (e.g., not placement connection),
* which doesn't set placements.
* In that case, the command sent could be either write or read, so we assume
* it is write to be on the safe side.
*/
return true;
}
dlist_foreach(placementIter, &connection->referencedPlacements)
{
ConnectionReference *connectionReference =

4
src/backend/distributed/connection/shared_connection_stats.c
View File

@ -136,9 +136,9 @@ PG_FUNCTION_INFO_V1(citus_remote_connection_stats);
Datum
citus_remote_connection_stats(PG_FUNCTION_ARGS)
{
TupleDesc tupleDescriptor = NULL;
CheckCitusVersion(ERROR);
TupleDesc tupleDescriptor = NULL;
Tuplestorestate *tupleStore = SetupTuplestore(fcinfo, &tupleDescriptor);
StoreAllRemoteConnectionStats(tupleStore, tupleDescriptor);

151
src/backend/distributed/deparser/citus_ruleutils.c
View File

@ -27,10 +27,12 @@
#include "catalog/indexing.h"
#include "catalog/namespace.h"
#include "catalog/pg_am.h"
#include "catalog/pg_attrdef.h"
#include "catalog/pg_attribute.h"
#include "catalog/pg_authid.h"
#include "catalog/pg_class.h"
#include "catalog/pg_collation.h"
#include "catalog/pg_depend.h"
#include "catalog/pg_extension.h"
#include "catalog/pg_foreign_data_wrapper.h"
#include "catalog/pg_index.h"
@ -40,9 +42,10 @@
#include "distributed/citus_ruleutils.h"
#include "distributed/listutils.h"
#include "distributed/multi_partitioning_utils.h"
#include "distributed/relay_utility.h"
#include "distributed/metadata_utility.h"
#include "distributed/metadata_cache.h"
#include "distributed/metadata_sync.h"
#include "distributed/metadata_utility.h"
#include "distributed/relay_utility.h"
#include "distributed/version_compat.h"
#include "foreign/foreign.h"
#include "lib/stringinfo.h"
@ -74,6 +77,8 @@ static void AppendStorageParametersToString(StringInfo stringBuffer,
List *optionList);
static void simple_quote_literal(StringInfo buf, const char *val);
static char * flatten_reloptions(Oid relid);
static Oid get_attrdef_oid(Oid relationId, AttrNumber attnum);
/*
* pg_get_extensiondef_string finds the foreign data wrapper that corresponds to
@ -365,6 +370,16 @@ pg_get_tableschemadef_string(Oid tableRelationId, bool includeSequenceDefaults,
appendStringInfo(&buffer, " DEFAULT %s", defaultString);
}
}
/*
* We should make sure that the type of the column that uses
* that sequence is supported
*/
if (contain_nextval_expression_walker(defaultNode, NULL))
{
EnsureSequenceTypeSupported(tableRelationId, defaultValue->adnum,
attributeForm->atttypid);
}
}
/* if this column has a not null constraint, append the constraint */
@ -483,6 +498,138 @@ pg_get_tableschemadef_string(Oid tableRelationId, bool includeSequenceDefaults,
}
/*
* EnsureSequenceTypeSupported ensures that the type of the column that uses
* a sequence on its DEFAULT is consistent with previous uses of the sequence (if any)
* It gets the AttrDefault OID from the given relationId and attnum, extracts the sequence
* id from it, and if any other distributed table uses that same sequence, it checks whether
* the types of the columns using the sequence match. If they don't, it errors out.
* Otherwise, the condition is ensured.
*/
void
EnsureSequenceTypeSupported(Oid relationId, AttrNumber attnum, Oid seqTypId)
{
/* get attrdefoid from the given relationId and attnum */
Oid attrdefOid = get_attrdef_oid(relationId, attnum);
/* retrieve the sequence id of the sequence found in nextval('seq') */
List *sequencesFromAttrDef = GetSequencesFromAttrDef(attrdefOid);
if (list_length(sequencesFromAttrDef) == 0)
{
/*
* We need this check because sometimes there are cases where the
* dependency between the table and the sequence is not formed
* One example is when the default is defined by
* DEFAULT nextval('seq_name'::text) (not by DEFAULT nextval('seq_name'))
* In these cases, sequencesFromAttrDef with be empty.
*/
return;
}
if (list_length(sequencesFromAttrDef) > 1)
{
/* to simplify and eliminate cases like "DEFAULT nextval('..') - nextval('..')" */
ereport(ERROR, (errmsg(
"More than one sequence in a column default"
" is not supported for distribution")));
}
Oid seqOid = lfirst_oid(list_head(sequencesFromAttrDef));
List *citusTableIdList = CitusTableTypeIdList(ANY_CITUS_TABLE_TYPE);
Oid citusTableId = InvalidOid;
foreach_oid(citusTableId, citusTableIdList)
{
List *attnumList = NIL;
List *dependentSequenceList = NIL;
GetDependentSequencesWithRelation(citusTableId, &attnumList,
&dependentSequenceList, 0);
ListCell *attnumCell = NULL;
ListCell *dependentSequenceCell = NULL;
forboth(attnumCell, attnumList, dependentSequenceCell,
dependentSequenceList)
{
AttrNumber currentAttnum = lfirst_int(attnumCell);
Oid currentSeqOid = lfirst_oid(dependentSequenceCell);
/*
* If another distributed table is using the same sequence
* in one of its column defaults, make sure the types of the
* columns match
*/
if (currentSeqOid == seqOid)
{
Oid currentSeqTypId = GetAttributeTypeOid(citusTableId,
currentAttnum);
if (seqTypId != currentSeqTypId)
{
char *sequenceName = generate_qualified_relation_name(
seqOid);
char *citusTableName =
generate_qualified_relation_name(citusTableId);
ereport(ERROR, (errmsg(
"The sequence %s is already used for a different"
" type in column %d of the table %s",
sequenceName, currentAttnum,
citusTableName)));
}
}
}
}
}
/*
* get_attrdef_oid gets the oid of the attrdef that has dependency with
* the given relationId (refobjid) and attnum (refobjsubid).
* If there is no such attrdef it returns InvalidOid.
* NOTE: we are iterating pg_depend here since this function is used together
* with other functions that iterate pg_depend. Normally, a look at pg_attrdef
* would make more sense.
*/
static Oid
get_attrdef_oid(Oid relationId, AttrNumber attnum)
{
Oid resultAttrdefOid = InvalidOid;
ScanKeyData key[3];
Relation depRel = table_open(DependRelationId, AccessShareLock);
ScanKeyInit(&key[0],
Anum_pg_depend_refclassid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationRelationId));
ScanKeyInit(&key[1],
Anum_pg_depend_refobjid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(relationId));
ScanKeyInit(&key[2],
Anum_pg_depend_refobjsubid,
BTEqualStrategyNumber, F_INT4EQ,
Int32GetDatum(attnum));
SysScanDesc scan = systable_beginscan(depRel, DependReferenceIndexId, true,
NULL, attnum ? 3 : 2, key);
HeapTuple tup;
while (HeapTupleIsValid(tup = systable_getnext(scan)))
{
Form_pg_depend deprec = (Form_pg_depend) GETSTRUCT(tup);
if (deprec->classid == AttrDefaultRelationId)
{
resultAttrdefOid = deprec->objid;
}
}
systable_endscan(scan);
table_close(depRel, AccessShareLock);
return resultAttrdefOid;
}
/*
* EnsureRelationKindSupported errors out if the given relation is not supported
* as a distributed relation.

49
src/backend/distributed/deparser/deparse_database_stmts.c Normal file
View File

@ -0,0 +1,49 @@
/*-------------------------------------------------------------------------
*
* deparse_database_stmts.c
*
* All routines to deparse database statements.
*
* Copyright (c), Citus Data, Inc.
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#include "catalog/namespace.h"
#include "lib/stringinfo.h"
#include "nodes/parsenodes.h"
#include "utils/builtins.h"
#include "distributed/citus_ruleutils.h"
#include "distributed/deparser.h"
static void AppendAlterDatabaseOwnerStmt(StringInfo buf, AlterOwnerStmt *stmt);
char *
DeparseAlterDatabaseOwnerStmt(Node *node)
{
AlterOwnerStmt *stmt = castNode(AlterOwnerStmt, node);
StringInfoData str = { 0 };
initStringInfo(&str);
Assert(stmt->objectType == OBJECT_DATABASE);
AppendAlterDatabaseOwnerStmt(&str, stmt);
return str.data;
}
static void
AppendAlterDatabaseOwnerStmt(StringInfo buf, AlterOwnerStmt *stmt)
{
Assert(stmt->objectType == OBJECT_DATABASE);
appendStringInfo(buf,
"ALTER DATABASE %s OWNER TO %s;",
quote_identifier(strVal((Value *) stmt->object)),
RoleSpecString(stmt->newowner, true));
}

158
src/backend/distributed/deparser/deparse_sequence_stmts.c Normal file
View File

@ -0,0 +1,158 @@
/*-------------------------------------------------------------------------
*
* deparse_sequence_stmts.c
*
* All routines to deparse sequence statements.
* This file contains all entry points specific for sequence statement
* deparsing
*
* Copyright (c), Citus Data, Inc.
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#include "catalog/namespace.h"
#include "distributed/deparser.h"
#include "utils/builtins.h"
#include "utils/lsyscache.h"
/* forward declaration for deparse functions */
static void AppendDropSequenceStmt(StringInfo buf, DropStmt *stmt);
static void AppendSequenceNameList(StringInfo buf, List *objects, ObjectType objtype);
static void AppendRenameSequenceStmt(StringInfo buf, RenameStmt *stmt);
/*
* DeparseDropSequenceStmt builds and returns a string representing the DropStmt
*/
char *
DeparseDropSequenceStmt(Node *node)
{
DropStmt *stmt = castNode(DropStmt, node);
StringInfoData str = { 0 };
initStringInfo(&str);
Assert(stmt->removeType == OBJECT_SEQUENCE);
AppendDropSequenceStmt(&str, stmt);
return str.data;
}
/*
* AppendDropSequenceStmt appends a string representing the DropStmt to a buffer
*/
static void
AppendDropSequenceStmt(StringInfo buf, DropStmt *stmt)
{
appendStringInfoString(buf, "DROP SEQUENCE ");
if (stmt->missing_ok)
{
appendStringInfoString(buf, "IF EXISTS ");
}
AppendSequenceNameList(buf, stmt->objects, stmt->removeType);
if (stmt->behavior == DROP_CASCADE)
{
appendStringInfoString(buf, " CASCADE");
}
appendStringInfoString(buf, ";");
}
/*
* AppendSequenceNameList appends a string representing the list of sequence names to a buffer
*/
static void
AppendSequenceNameList(StringInfo buf, List *objects, ObjectType objtype)
{
ListCell *objectCell = NULL;
foreach(objectCell, objects)
{
if (objectCell != list_head(objects))
{
appendStringInfo(buf, ", ");
}
RangeVar *seq = makeRangeVarFromNameList((List *) lfirst(objectCell));
if (seq->schemaname == NULL)
{
Oid schemaOid = RangeVarGetCreationNamespace(seq);
seq->schemaname = get_namespace_name(schemaOid);
}
char *qualifiedSequenceName = quote_qualified_identifier(seq->schemaname,
seq->relname);
appendStringInfoString(buf, qualifiedSequenceName);
}
}
/*
* DeparseRenameSequenceStmt builds and returns a string representing the RenameStmt
*/
char *
DeparseRenameSequenceStmt(Node *node)
{
RenameStmt *stmt = castNode(RenameStmt, node);
StringInfoData str = { 0 };
initStringInfo(&str);
Assert(stmt->renameType == OBJECT_SEQUENCE);
AppendRenameSequenceStmt(&str, stmt);
return str.data;
}
/*
* AppendRenameSequenceStmt appends a string representing the RenameStmt to a buffer
*/
static void
AppendRenameSequenceStmt(StringInfo buf, RenameStmt *stmt)
{
RangeVar *seq = stmt->relation;
char *qualifiedSequenceName = quote_qualified_identifier(seq->schemaname,
seq->relname);
appendStringInfoString(buf, "ALTER SEQUENCE ");
if (stmt->missing_ok)
{
appendStringInfoString(buf, "IF EXISTS ");
}
appendStringInfoString(buf, qualifiedSequenceName);
appendStringInfo(buf, " RENAME TO %s", quote_identifier(stmt->newname));
}
/*
* QualifyRenameSequenceStmt transforms a
* ALTER SEQUENCE .. RENAME TO ..
* statement in place and makes the sequence name fully qualified.
*/
void
QualifyRenameSequenceStmt(Node *node)
{
RenameStmt *stmt = castNode(RenameStmt, node);
Assert(stmt->renameType == OBJECT_SEQUENCE);
RangeVar *seq = stmt->relation;
if (seq->schemaname == NULL)
{
Oid schemaOid = RangeVarGetCreationNamespace(seq);
seq->schemaname = get_namespace_name(schemaOid);
}
}

130
src/backend/distributed/deparser/deparse_table_stmts.c
View File

@ -14,9 +14,13 @@
#include "distributed/deparser.h"
#include "nodes/nodes.h"
#include "nodes/parsenodes.h"
#include "parser/parse_type.h"
#include "utils/builtins.h"
static void AppendAlterTableSchemaStmt(StringInfo buf, AlterObjectSchemaStmt *stmt);
static void AppendAlterTableStmt(StringInfo buf, AlterTableStmt *stmt);
static void AppendAlterTableCmd(StringInfo buf, AlterTableCmd *alterTableCmd);
static void AppendAlterTableCmdAddColumn(StringInfo buf, AlterTableCmd *alterTableCmd);
char *
DeparseAlterTableSchemaStmt(Node *node)
@ -46,3 +50,129 @@ AppendAlterTableSchemaStmt(StringInfo buf, AlterObjectSchemaStmt *stmt)
const char *newSchemaName = quote_identifier(stmt->newschema);
appendStringInfo(buf, "%s SET SCHEMA %s;", tableName, newSchemaName);
}
/*
* DeparseAlterTableStmt builds and returns a string representing the
* AlterTableStmt where the object acted upon is of kind OBJECT_TABLE
*/
char *
DeparseAlterTableStmt(Node *node)
{
AlterTableStmt *stmt = castNode(AlterTableStmt, node);
StringInfoData str = { 0 };
initStringInfo(&str);
Assert(stmt->relkind == OBJECT_TABLE);
AppendAlterTableStmt(&str, stmt);
return str.data;
}
/*
* AppendAlterTableStmt builds and returns an SQL command representing an
* ALTER TABLE statement from given AlterTableStmt object where the object
* acted upon is of kind OBJECT_TABLE
*/
static void
AppendAlterTableStmt(StringInfo buf, AlterTableStmt *stmt)
{
const char *identifier = quote_qualified_identifier(stmt->relation->schemaname,
stmt->relation->relname);
ListCell *cmdCell = NULL;
Assert(stmt->relkind == OBJECT_TABLE);
appendStringInfo(buf, "ALTER TABLE %s", identifier);
foreach(cmdCell, stmt->cmds)
{
if (cmdCell != list_head(stmt->cmds))
{
appendStringInfoString(buf, ", ");
}
AlterTableCmd *alterTableCmd = castNode(AlterTableCmd, lfirst(cmdCell));
AppendAlterTableCmd(buf, alterTableCmd);
}
appendStringInfoString(buf, ";");
}
/*
* AppendAlterTableCmd builds and appends to the given buffer a command
* from given AlterTableCmd object. Currently supported commands are of type
* AT_AddColumn and AT_SetNotNull
*/
static void
AppendAlterTableCmd(StringInfo buf, AlterTableCmd *alterTableCmd)
{
switch (alterTableCmd->subtype)
{
case AT_AddColumn:
{
AppendAlterTableCmdAddColumn(buf, alterTableCmd);
break;
}
default:
{
ereport(ERROR, (errmsg("unsupported subtype for alter table command"),
errdetail("sub command type: %d", alterTableCmd->subtype)));
}
}
}
/*
* AppendAlterTableCmd builds and appends to the given buffer an AT_AddColumn command
* from given AlterTableCmd object in the form ADD COLUMN ...
*/
static void
AppendAlterTableCmdAddColumn(StringInfo buf, AlterTableCmd *alterTableCmd)
{
Assert(alterTableCmd->subtype == AT_AddColumn);
appendStringInfoString(buf, " ADD COLUMN ");
ColumnDef *columnDefinition = (ColumnDef *) alterTableCmd->def;
/*
* the way we use the deparser now, constraints are always NULL
* adding this check for ColumnDef consistency
*/
if (columnDefinition->constraints != NULL)
{
ereport(ERROR, (errmsg("Constraints are not supported for AT_AddColumn")));
}
if (columnDefinition->colname)
{
appendStringInfo(buf, "%s ", quote_identifier(columnDefinition->colname));
}
int32 typmod = 0;
Oid typeOid = InvalidOid;
bits16 formatFlags = FORMAT_TYPE_TYPEMOD_GIVEN | FORMAT_TYPE_FORCE_QUALIFY;
typenameTypeIdAndMod(NULL, columnDefinition->typeName, &typeOid, &typmod);
appendStringInfo(buf, "%s", format_type_extended(typeOid, typmod,
formatFlags));
if (columnDefinition->is_not_null)
{
appendStringInfoString(buf, " NOT NULL");
}
/*
* the way we use the deparser now, collation is never used
* since the data type of columns that use sequences for default
* are only int,smallint and bigint (never text, varchar, char)
* Adding this part only for ColumnDef consistency
*/
Oid collationOid = GetColumnDefCollation(NULL, columnDefinition, typeOid);
if (OidIsValid(collationOid))
{
const char *identifier = FormatCollateBEQualified(collationOid);
appendStringInfo(buf, " COLLATE %s", identifier);
}
}

396
src/backend/distributed/executor/adaptive_executor.c
View File

@ -124,6 +124,7 @@
#include "miscadmin.h"
#include "pgstat.h"
#include <math.h>
#include <sys/stat.h>
#include <unistd.h>
@ -142,6 +143,7 @@
#include "distributed/deparse_shard_query.h"
#include "distributed/shared_connection_stats.h"
#include "distributed/distributed_execution_locks.h"
#include "distributed/intermediate_result_pruning.h"
#include "distributed/listutils.h"
#include "distributed/local_executor.h"
#include "distributed/multi_client_executor.h"
@ -413,6 +415,10 @@ typedef struct WorkerPool
* use it anymore.
*/
WorkerPoolFailureState failureState;
/* execution statistics per pool, in microseconds */
uint64 totalTaskExecutionTime;
int totalExecutedTasks;
} WorkerPool;
struct TaskPlacementExecution;
@ -472,6 +478,8 @@ bool EnableBinaryProtocol = false;
/* GUC, number of ms to wait between opening connections to the same worker */
int ExecutorSlowStartInterval = 10;
bool EnableCostBasedConnectionEstablishment = true;
bool PreventIncompleteConnectionEstablishment = true;
/*
@ -634,6 +642,12 @@ static WorkerSession * FindOrCreateWorkerSession(WorkerPool *workerPool,
static void ManageWorkerPool(WorkerPool *workerPool);
static bool ShouldWaitForSlowStart(WorkerPool *workerPool);
static int CalculateNewConnectionCount(WorkerPool *workerPool);
static bool UsingExistingSessionsCheaperThanEstablishingNewConnections(int
readyTaskCount,
WorkerPool *
workerPool);
static double AvgTaskExecutionTimeApproximation(WorkerPool *workerPool);
static double AvgConnectionEstablishmentTime(WorkerPool *workerPool);
static void OpenNewConnections(WorkerPool *workerPool, int newConnectionCount,
TransactionProperties *transactionProperties);
static void CheckConnectionTimeout(WorkerPool *workerPool);
@ -650,6 +664,7 @@ static bool StartPlacementExecutionOnSession(TaskPlacementExecution *placementEx
static bool SendNextQuery(TaskPlacementExecution *placementExecution,
WorkerSession *session);
static void ConnectionStateMachine(WorkerSession *session);
static bool HasUnfinishedTaskForSession(WorkerSession *session);
static void HandleMultiConnectionSuccess(WorkerSession *session);
static bool HasAnyConnectionFailure(WorkerPool *workerPool);
static void Activate2PCIfModifyingTransactionExpandsToNewNode(WorkerSession *session);
@ -675,15 +690,20 @@ static void ExtractParametersForRemoteExecution(ParamListInfo paramListInfo,
Oid **parameterTypes,
const char ***parameterValues);
static int GetEventSetSize(List *sessionList);
static bool HasIncompleteConnectionEstablishment(DistributedExecution *execution);
static int RebuildWaitEventSet(DistributedExecution *execution);
static void ProcessWaitEvents(DistributedExecution *execution, WaitEvent *events, int
eventCount, bool *cancellationReceived);
static long MillisecondsBetweenTimestamps(instr_time startTime, instr_time endTime);
static uint64 MicrosecondsBetweenTimestamps(instr_time startTime, instr_time endTime);
static HeapTuple BuildTupleFromBytes(AttInMetadata *attinmeta, fmStringInfo *values);
static AttInMetadata * TupleDescGetAttBinaryInMetadata(TupleDesc tupdesc);
static int WorkerPoolCompare(const void *lhsKey, const void *rhsKey);
static void SetAttributeInputMetadata(DistributedExecution *execution,
ShardCommandExecution *shardCommandExecution);
static void LookupTaskPlacementHostAndPort(ShardPlacement *taskPlacement, char **nodeName,
int *nodePort);
static bool IsDummyPlacement(ShardPlacement *taskPlacement);
/*
* AdaptiveExecutorPreExecutorRun gets called right before postgres starts its executor
@ -742,6 +762,12 @@ AdaptiveExecutor(CitusScanState *scanState)
/* we should only call this once before the scan finished */
Assert(!scanState->finishedRemoteScan);
MemoryContext localContext = AllocSetContextCreate(CurrentMemoryContext,
"AdaptiveExecutor",
ALLOCSET_DEFAULT_SIZES);
MemoryContext oldContext = MemoryContextSwitchTo(localContext);
/* Reset Task fields that are only valid for a single execution */
ResetExplainAnalyzeData(taskList);
@ -830,6 +856,8 @@ AdaptiveExecutor(CitusScanState *scanState)
SortTupleStore(scanState);
}
MemoryContextSwitchTo(oldContext);
return resultSlot;
}
@ -1227,7 +1255,7 @@ StartDistributedExecution(DistributedExecution *execution)
if (xactProperties->requires2PC)
{
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
}
/*
@ -1630,8 +1658,10 @@ CleanUpSessions(DistributedExecution *execution)
{
MultiConnection *connection = session->connection;
ereport(DEBUG4, (errmsg("Total number of commands sent over the session %ld: %ld",
session->sessionId, session->commandsSent)));
ereport(DEBUG4, (errmsg("Total number of commands sent over the session %ld: %ld "
"to node %s:%d", session->sessionId,
session->commandsSent,
connection->hostname, connection->port)));
UnclaimConnection(connection);
@ -1751,8 +1781,10 @@ AssignTasksToConnectionsOrWorkerPool(DistributedExecution *execution)
foreach_ptr(taskPlacement, task->taskPlacementList)
{
int connectionFlags = 0;
char *nodeName = taskPlacement->nodeName;
int nodePort = taskPlacement->nodePort;
char *nodeName = NULL;
int nodePort = 0;
LookupTaskPlacementHostAndPort(taskPlacement, &nodeName, &nodePort);
WorkerPool *workerPool = FindOrCreateWorkerPool(execution, nodeName,
nodePort);
@ -1900,6 +1932,48 @@ AssignTasksToConnectionsOrWorkerPool(DistributedExecution *execution)
}
/*
* LookupTaskPlacementHostAndPort sets the nodename and nodeport for the given task placement
* with a lookup.
*/
static void
LookupTaskPlacementHostAndPort(ShardPlacement *taskPlacement, char **nodeName,
int *nodePort)
{
if (IsDummyPlacement(taskPlacement))
{
/*
* If we create a dummy placement for the local node, it is possible
* that the entry doesn't exist in pg_dist_node, hence a lookup will fail.
* In that case we want to use the dummy placements values.
*/
*nodeName = taskPlacement->nodeName;
*nodePort = taskPlacement->nodePort;
}
else
{
/*
* We want to lookup the node information again since it is possible that
* there were changes in pg_dist_node and we will get those invalidations
* in LookupNodeForGroup.
*/
WorkerNode *workerNode = LookupNodeForGroup(taskPlacement->groupId);
*nodeName = workerNode->workerName;
*nodePort = workerNode->workerPort;
}
}
/*
* IsDummyPlacement returns true if the given placement is a dummy placement.
*/
static bool
IsDummyPlacement(ShardPlacement *taskPlacement)
{
return taskPlacement->nodeId == LOCAL_NODE_ID;
}
/*
* WorkerPoolCompare is based on WorkerNodeCompare function. The function
* compares two worker nodes by their host name and port number.
@ -2217,7 +2291,26 @@ RunDistributedExecution(DistributedExecution *execution)
/* always (re)build the wait event set the first time */
execution->rebuildWaitEventSet = true;
while (execution->unfinishedTaskCount > 0 && !cancellationReceived)
/*
* Iterate until all the tasks are finished. Once all the tasks
* are finished, ensure that that all the connection initializations
* are also finished. Otherwise, those connections are terminated
* abruptly before they are established (or failed). Instead, we let
* the ConnectionStateMachine() to properly handle them.
*
* Note that we could have the connections that are not established
* as a side effect of slow-start algorithm. At the time the algorithm
* decides to establish new connections, the execution might have tasks
* to finish. But, the execution might finish before the new connections
* are established.
*
* Note that the rules explained above could be overriden by any
* cancellation to the query. In that case, we terminate the execution
* irrespective of the current status of the tasks or the connections.
*/
while (!cancellationReceived &&
(execution->unfinishedTaskCount > 0 ||
HasIncompleteConnectionEstablishment(execution)))
{
WorkerPool *workerPool = NULL;
foreach_ptr(workerPool, execution->workerList)
@ -2299,6 +2392,33 @@ RunDistributedExecution(DistributedExecution *execution)
}
/*
* HasIncompleteConnectionEstablishment returns true if any of the connections
* that has been initiated by the executor is in initilization stage.
*/
static bool
HasIncompleteConnectionEstablishment(DistributedExecution *execution)
{
if (!PreventIncompleteConnectionEstablishment)
{
return false;
}
WorkerSession *session = NULL;
foreach_ptr(session, execution->sessionList)
{
MultiConnection *connection = session->connection;
if (connection->connectionState == MULTI_CONNECTION_INITIAL ||
connection->connectionState == MULTI_CONNECTION_CONNECTING)
{
return true;
}
}
return false;
}
/*
* RebuildWaitEventSet updates the waitEventSet for the distributed execution.
* This happens when the connection set for the distributed execution is changed,
@ -2398,6 +2518,9 @@ ManageWorkerPool(WorkerPool *workerPool)
return;
}
/* increase the open rate every cycle (like TCP slow start) */
workerPool->maxNewConnectionsPerCycle += 1;
OpenNewConnections(workerPool, newConnectionCount, execution->transactionProperties);
/*
@ -2584,16 +2707,176 @@ CalculateNewConnectionCount(WorkerPool *workerPool)
* than the target pool size.
*/
newConnectionCount = Min(newConnectionsForReadyTasks, maxNewConnectionCount);
if (newConnectionCount > 0)
if (EnableCostBasedConnectionEstablishment && newConnectionCount > 0 &&
initiatedConnectionCount <= MaxCachedConnectionsPerWorker &&
UsingExistingSessionsCheaperThanEstablishingNewConnections(
readyTaskCount, workerPool))
{
/* increase the open rate every cycle (like TCP slow start) */
workerPool->maxNewConnectionsPerCycle += 1;
/*
* Before giving the decision, we do one more check. If the cost of
* executing the remaining tasks over the existing sessions in the
* pool is cheaper than establishing new connections and executing
* the tasks over the new connections, we prefer the former.
*
* For cached connections we should ignore any optimizations as
* cached connections are almost free to get. In other words,
* as long as there are cached connections that the pool has
* not used yet, aggressively use these already established
* connections.
*
* Note that until MaxCachedConnectionsPerWorker has already been
* established within the session, we still need to establish
* the connections right now.
*
* Also remember that we are not trying to find the optimal number
* of connections for the remaining tasks here. Our goal is to prevent
* connection establishments that are absolutely unnecessary. In the
* future, we may improve the calculations below to find the optimal
* number of new connections required.
*/
return 0;
}
}
return newConnectionCount;
}
/*
* UsingExistingSessionsCheaperThanEstablishingNewConnections returns true if
* using the already established connections takes less time compared to opening
* new connections based on the current execution's stats.
*
* The function returns false if the current execution has not established any connections
* or finished any tasks (e.g., no stats to act on).
*/
static bool
UsingExistingSessionsCheaperThanEstablishingNewConnections(int readyTaskCount,
WorkerPool *workerPool)
{
int activeConnectionCount = workerPool->activeConnectionCount;
if (workerPool->totalExecutedTasks < 1 || activeConnectionCount < 1)
{
/*
* The pool has not finished any connection establishment or
* task yet. So, we refrain from optimizing the execution.
*/
return false;
}
double avgTaskExecutionTime = AvgTaskExecutionTimeApproximation(workerPool);
double avgConnectionEstablishmentTime = AvgConnectionEstablishmentTime(workerPool);
/* we assume that we are halfway through the execution */
double remainingTimeForActiveTaskExecutionsToFinish = avgTaskExecutionTime / 2;
/*
* We use "newConnectionCount" as if it is the task count as
* we are only interested in this iteration of CalculateNewConnectionCount().
*/
double totalTimeToExecuteNewTasks = avgTaskExecutionTime * readyTaskCount;
double estimatedExecutionTimeForNewTasks =
floor(totalTimeToExecuteNewTasks / activeConnectionCount);
/*
* First finish the already running tasks, and then use the connections
* to execute the new tasks.
*/
double costOfExecutingTheTasksOverExistingConnections =
remainingTimeForActiveTaskExecutionsToFinish +
estimatedExecutionTimeForNewTasks;
/*
* For every task, the executor is supposed to establish one
* connection and then execute the task over the connection.
*/
double costOfExecutingTheTasksOverNewConnection =
(avgTaskExecutionTime + avgConnectionEstablishmentTime);
return (costOfExecutingTheTasksOverExistingConnections <=
costOfExecutingTheTasksOverNewConnection);
}
/*
* AvgTaskExecutionTimeApproximation returns the approximation of the average task
* execution times on the workerPool.
*/
static double
AvgTaskExecutionTimeApproximation(WorkerPool *workerPool)
{
uint64 totalTaskExecutionTime = workerPool->totalTaskExecutionTime;
int taskCount = workerPool->totalExecutedTasks;
instr_time now;
INSTR_TIME_SET_CURRENT(now);
WorkerSession *session = NULL;
foreach_ptr(session, workerPool->sessionList)
{
/*
* Involve the tasks that are currently running. We do this to
* make sure that the execution responds with new connections
* quickly if the actively running tasks
*/
TaskPlacementExecution *placementExecution = session->currentTask;
if (placementExecution != NULL &&
placementExecution->executionState == PLACEMENT_EXECUTION_RUNNING)
{
uint64 durationInMicroSecs =
MicrosecondsBetweenTimestamps(placementExecution->startTime, now);
/*
* Our approximation is that we assume that the task execution is
* just in the halfway through.
*/
totalTaskExecutionTime += (2 * durationInMicroSecs);
taskCount += 1;
}
}
return taskCount == 0 ? 0 : ((double) totalTaskExecutionTime / taskCount);
}
/*
* AvgConnectionEstablishmentTime calculates the average connection establishment times
* for the input workerPool.
*/
static double
AvgConnectionEstablishmentTime(WorkerPool *workerPool)
{
double totalTimeMicrosec = 0;
int sessionCount = 0;
WorkerSession *session = NULL;
foreach_ptr(session, workerPool->sessionList)
{
MultiConnection *connection = session->connection;
/*
* There could be MaxCachedConnectionsPerWorker connections that are
* already connected. Those connections might skew the average
* connection establishment times for the current execution. The reason
* is that they are established earlier and the connection establishment
* times might be different at the moment those connections are established.
*/
if (connection->connectionState == MULTI_CONNECTION_CONNECTED)
{
long connectionEstablishmentTime =
MicrosecondsBetweenTimestamps(connection->connectionEstablishmentStart,
connection->connectionEstablishmentEnd);
totalTimeMicrosec += connectionEstablishmentTime;
++sessionCount;
}
}
return (sessionCount == 0) ? 0 : (totalTimeMicrosec / sessionCount);
}
/*
* OpenNewConnections opens the given amount of connections for the given workerPool.
*/
@ -2898,6 +3181,18 @@ MillisecondsBetweenTimestamps(instr_time startTime, instr_time endTime)
}
/*
* MicrosecondsBetweenTimestamps is a helper to get the number of microseconds
* between timestamps.
*/
static uint64
MicrosecondsBetweenTimestamps(instr_time startTime, instr_time endTime)
{
INSTR_TIME_SUBTRACT(endTime, startTime);
return INSTR_TIME_GET_MICROSEC(endTime);
}
/*
* ConnectionStateMachine opens a connection and descends into the transaction
* state machine when ready.
@ -2995,8 +3290,32 @@ ConnectionStateMachine(WorkerSession *session)
case MULTI_CONNECTION_CONNECTED:
{
/* connection is ready, run the transaction state machine */
if (HasUnfinishedTaskForSession(session))
{
/*
* Connection is ready, and we have unfinished tasks.
* So, run the transaction state machine.
*/
TransactionStateMachine(session);
}
else
{
/*
* Connection is ready, but we don't have any unfinished
* tasks that this session can execute.
*
* Note that we can be in a situation where the executor
* decides to establish a connection, but not need to
* use it at the time the connection is established. This could
* happen when the earlier connections manages to finish all the
* tasks after this connection
*
* As no tasks are ready to be executed at the moment, we
* mark the socket readable to get any notices if exists.
*/
UpdateConnectionWaitFlags(session, WL_SOCKET_READABLE);
}
break;
}
@ -3111,6 +3430,41 @@ ConnectionStateMachine(WorkerSession *session)
}
/*
* HasUnfinishedTaskForSession gets a session and returns true if there
* are any tasks that this session can execute.
*/
static bool
HasUnfinishedTaskForSession(WorkerSession *session)
{
if (session->currentTask != NULL)
{
/* the session is executing a command right now */
return true;
}
dlist_head *sessionReadyTaskQueue = &(session->readyTaskQueue);
if (!dlist_is_empty(sessionReadyTaskQueue))
{
/* session has an assigned task, which is ready for execution */
return true;
}
WorkerPool *workerPool = session->workerPool;
dlist_head *poolReadyTaskQueue = &(workerPool->readyTaskQueue);
if (!dlist_is_empty(poolReadyTaskQueue))
{
/*
* Pool has unassigned tasks that can be executed
* by the input session.
*/
return true;
}
return false;
}
/*
* HandleMultiConnectionSuccess logs the established connection and updates
* connection's state.
@ -3124,10 +3478,10 @@ HandleMultiConnectionSuccess(WorkerSession *session)
MarkConnectionConnected(connection);
ereport(DEBUG4, (errmsg("established connection to %s:%d for "
"session %ld in %ld msecs",
"session %ld in %ld microseconds",
connection->hostname, connection->port,
session->sessionId,
MillisecondsBetweenTimestamps(
MicrosecondsBetweenTimestamps(
connection->connectionEstablishmentStart,
connection->connectionEstablishmentEnd))));
@ -3188,7 +3542,7 @@ Activate2PCIfModifyingTransactionExpandsToNewNode(WorkerSession *session)
* just opened, which means we're now going to make modifications
* over multiple connections. Activate 2PC!
*/
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
}
}
@ -4270,19 +4624,20 @@ PlacementExecutionDone(TaskPlacementExecution *placementExecution, bool succeede
Assert(INSTR_TIME_IS_ZERO(placementExecution->endTime));
INSTR_TIME_SET_CURRENT(placementExecution->endTime);
uint64 durationMicrosecs =
MicrosecondsBetweenTimestamps(placementExecution->startTime,
placementExecution->endTime);
workerPool->totalTaskExecutionTime += durationMicrosecs;
workerPool->totalExecutedTasks += 1;
if (IsLoggableLevel(DEBUG4))
{
long durationMillisecs =
MillisecondsBetweenTimestamps(placementExecution->startTime,
placementExecution->endTime);
ereport(DEBUG4, (errmsg("task execution (%d) for placement (%ld) on anchor "
"shard (%ld) finished in %ld msecs on worker "
"shard (%ld) finished in %ld microseconds on worker "
"node %s:%d", shardCommandExecution->task->taskId,
placementExecution->shardPlacement->placementId,
shardCommandExecution->task->anchorShardId,
durationMillisecs, workerPool->nodeName,
durationMicrosecs, workerPool->nodeName,
workerPool->nodePort)));
}
}
@ -4457,8 +4812,6 @@ ScheduleNextPlacementExecution(TaskPlacementExecution *placementExecution, bool
executionOrder == EXECUTION_ORDER_SEQUENTIAL)
{
TaskPlacementExecution *nextPlacementExecution = NULL;
int placementExecutionCount PG_USED_FOR_ASSERTS_ONLY =
shardCommandExecution->placementExecutionCount;
/* find a placement execution that is not yet marked as failed */
do {
@ -4469,6 +4822,7 @@ ScheduleNextPlacementExecution(TaskPlacementExecution *placementExecution, bool
* If all tasks failed then we should already have errored out.
* Still, be defensive and throw error instead of crashes.
*/
int placementExecutionCount = shardCommandExecution->placementExecutionCount;
if (nextPlacementExecutionIndex >= placementExecutionCount)
{
WorkerPool *workerPool = placementExecution->workerPool;

13
src/backend/distributed/executor/citus_custom_scan.c
View File

@ -189,6 +189,12 @@ CitusBeginScan(CustomScanState *node, EState *estate, int eflags)
{
CitusBeginModifyScan(node, estate, eflags);
}
/*
* In case of a prepared statement, we will see this distributed plan again
* on the next execution with a higher usage counter.
*/
distributedPlan->numberOfTimesExecuted++;
}
@ -315,6 +321,11 @@ CitusBeginModifyScan(CustomScanState *node, EState *estate, int eflags)
PlanState *planState = &(scanState->customScanState.ss.ps);
DistributedPlan *originalDistributedPlan = scanState->distributedPlan;
MemoryContext localContext = AllocSetContextCreate(CurrentMemoryContext,
"CitusBeginModifyScan",
ALLOCSET_DEFAULT_SIZES);
MemoryContext oldContext = MemoryContextSwitchTo(localContext);
DistributedPlan *currentPlan =
CopyDistributedPlanWithoutCache(originalDistributedPlan);
scanState->distributedPlan = currentPlan;
@ -405,6 +416,8 @@ CitusBeginModifyScan(CustomScanState *node, EState *estate, int eflags)
*/
CacheLocalPlanForShardQuery(task, originalDistributedPlan);
}
MemoryContextSwitchTo(oldContext);
}

20
src/backend/distributed/executor/intermediate_results.c
View File

@ -118,6 +118,8 @@ PG_FUNCTION_INFO_V1(fetch_intermediate_results);
Datum
broadcast_intermediate_result(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *resultIdText = PG_GETARG_TEXT_P(0);
char *resultIdString = text_to_cstring(resultIdText);
text *queryText = PG_GETARG_TEXT_P(1);
@ -125,8 +127,6 @@ broadcast_intermediate_result(PG_FUNCTION_ARGS)
bool writeLocalFile = false;
ParamListInfo paramListInfo = NULL;
CheckCitusVersion(ERROR);
/*
* Make sure that this transaction has a distributed transaction ID.
*
@ -159,6 +159,8 @@ broadcast_intermediate_result(PG_FUNCTION_ARGS)
Datum
create_intermediate_result(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *resultIdText = PG_GETARG_TEXT_P(0);
char *resultIdString = text_to_cstring(resultIdText);
text *queryText = PG_GETARG_TEXT_P(1);
@ -167,8 +169,6 @@ create_intermediate_result(PG_FUNCTION_ARGS)
bool writeLocalFile = true;
ParamListInfo paramListInfo = NULL;
CheckCitusVersion(ERROR);
/*
* Make sure that this transaction has a distributed transaction ID.
*
@ -771,13 +771,13 @@ IntermediateResultSize(const char *resultId)
Datum
read_intermediate_result(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Datum resultId = PG_GETARG_DATUM(0);
Datum copyFormatOidDatum = PG_GETARG_DATUM(1);
Datum copyFormatLabelDatum = DirectFunctionCall1(enum_out, copyFormatOidDatum);
char *copyFormatLabel = DatumGetCString(copyFormatLabelDatum);
CheckCitusVersion(ERROR);
ReadIntermediateResultsIntoFuncOutput(fcinfo, copyFormatLabel, &resultId, 1);
PG_RETURN_DATUM(0);
@ -794,14 +794,14 @@ read_intermediate_result(PG_FUNCTION_ARGS)
Datum
read_intermediate_result_array(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
ArrayType *resultIdObject = PG_GETARG_ARRAYTYPE_P(0);
Datum copyFormatOidDatum = PG_GETARG_DATUM(1);
Datum copyFormatLabelDatum = DirectFunctionCall1(enum_out, copyFormatOidDatum);
char *copyFormatLabel = DatumGetCString(copyFormatLabelDatum);
CheckCitusVersion(ERROR);
int32 resultCount = ArrayGetNItems(ARR_NDIM(resultIdObject), ARR_DIMS(
resultIdObject));
Datum *resultIdArray = DeconstructArrayObject(resultIdObject);
@ -874,6 +874,8 @@ ReadIntermediateResultsIntoFuncOutput(FunctionCallInfo fcinfo, char *copyFormat,
Datum
fetch_intermediate_results(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
ArrayType *resultIdObject = PG_GETARG_ARRAYTYPE_P(0);
Datum *resultIdArray = DeconstructArrayObject(resultIdObject);
int32 resultCount = ArrayObjectCount(resultIdObject);
@ -885,8 +887,6 @@ fetch_intermediate_results(PG_FUNCTION_ARGS)
int resultIndex = 0;
int64 totalBytesWritten = 0L;
CheckCitusVersion(ERROR);
if (resultCount == 0)
{
PG_RETURN_INT64(0);

73
src/backend/distributed/executor/local_executor.c
View File

@ -108,8 +108,15 @@
bool EnableLocalExecution = true;
bool LogLocalCommands = false;
int LocalExecutorLevel = 0;
static LocalExecutionStatus CurrentLocalExecutionStatus = LOCAL_EXECUTION_OPTIONAL;
static uint64 ExecuteLocalTaskListInternal(List *taskList,
ParamListInfo paramListInfo,
DistributedPlan *distributedPlan,
TupleDestination *defaultTupleDest,
bool isUtilityCommand);
static void SplitLocalAndRemotePlacements(List *taskPlacementList,
List **localTaskPlacementList,
List **remoteTaskPlacementList);
@ -200,10 +207,8 @@ ExecuteLocalTaskListExtended(List *taskList,
TupleDestination *defaultTupleDest,
bool isUtilityCommand)
{
ParamListInfo paramListInfo = copyParamList(orig_paramListInfo);
int numParams = 0;
Oid *parameterTypes = NULL;
uint64 totalRowsProcessed = 0;
ParamListInfo paramListInfo = copyParamList(orig_paramListInfo);
/*
* Even if we are executing local tasks, we still enable
@ -218,6 +223,38 @@ ExecuteLocalTaskListExtended(List *taskList,
*/
UseCoordinatedTransaction();
LocalExecutorLevel++;
PG_TRY();
{
totalRowsProcessed = ExecuteLocalTaskListInternal(taskList, paramListInfo,
distributedPlan,
defaultTupleDest,
isUtilityCommand);
}
PG_CATCH();
{
LocalExecutorLevel--;
PG_RE_THROW();
}
PG_END_TRY();
LocalExecutorLevel--;
return totalRowsProcessed;
}
static uint64
ExecuteLocalTaskListInternal(List *taskList,
ParamListInfo paramListInfo,
DistributedPlan *distributedPlan,
TupleDestination *defaultTupleDest,
bool isUtilityCommand)
{
uint64 totalRowsProcessed = 0;
int numParams = 0;
Oid *parameterTypes = NULL;
if (paramListInfo != NULL)
{
/* not used anywhere, so declare here */
@ -229,9 +266,19 @@ ExecuteLocalTaskListExtended(List *taskList,
numParams = paramListInfo->numParams;
}
/*
* Use a new memory context that gets reset after every task to free
* the deparsed query string and query plan.
*/
MemoryContext loopContext = AllocSetContextCreate(CurrentMemoryContext,
"ExecuteLocalTaskListExtended",
ALLOCSET_DEFAULT_SIZES);
Task *task = NULL;
foreach_ptr(task, taskList)
{
MemoryContext oldContext = MemoryContextSwitchTo(loopContext);
TupleDestination *tupleDest = task->tupleDest ?
task->tupleDest :
defaultTupleDest;
@ -253,7 +300,7 @@ ExecuteLocalTaskListExtended(List *taskList,
* queries are also ReadOnly, our 2PC logic is smart enough to skip sending
* PREPARE to those connections.
*/
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
}
LogLocalCommand(task);
@ -261,6 +308,9 @@ ExecuteLocalTaskListExtended(List *taskList,
if (isUtilityCommand)
{
ExecuteUtilityCommand(TaskQueryString(task));
MemoryContextSwitchTo(oldContext);
MemoryContextReset(loopContext);
continue;
}
@ -308,6 +358,9 @@ ExecuteLocalTaskListExtended(List *taskList,
totalRowsProcessed +=
LocallyPlanAndExecuteMultipleQueries(queryStringList, tupleDest,
task);
MemoryContextSwitchTo(oldContext);
MemoryContextReset(loopContext);
continue;
}
@ -343,6 +396,9 @@ ExecuteLocalTaskListExtended(List *taskList,
totalRowsProcessed +=
ExecuteLocalTaskPlan(localPlan, shardQueryString,
tupleDest, task, paramListInfo);
MemoryContextSwitchTo(oldContext);
MemoryContextReset(loopContext);
}
return totalRowsProcessed;
@ -582,6 +638,12 @@ ExecuteLocalTaskPlan(PlannedStmt *taskPlan, char *queryString,
RecordNonDistTableAccessesForTask(task);
MemoryContext localContext = AllocSetContextCreate(CurrentMemoryContext,
"ExecuteLocalTaskPlan",
ALLOCSET_DEFAULT_SIZES);
MemoryContext oldContext = MemoryContextSwitchTo(localContext);
/*
* Some tuple destinations look at task->taskPlacementList to determine
* where the result came from using the placement index. Since a local
@ -625,6 +687,9 @@ ExecuteLocalTaskPlan(PlannedStmt *taskPlan, char *queryString,
FreeQueryDesc(queryDesc);
MemoryContextSwitchTo(oldContext);
MemoryContextDelete(localContext);
return totalRowsProcessed;
}

4
src/backend/distributed/executor/partitioned_intermediate_results.c
View File

@ -107,6 +107,8 @@ PG_FUNCTION_INFO_V1(worker_partition_query_result);
Datum
worker_partition_query_result(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
ReturnSetInfo *resultInfo = (ReturnSetInfo *) fcinfo->resultinfo;
text *resultIdPrefixText = PG_GETARG_TEXT_P(0);
@ -136,8 +138,6 @@ worker_partition_query_result(PG_FUNCTION_ARGS)
bool binaryCopy = PG_GETARG_BOOL(6);
CheckCitusVersion(ERROR);
if (!IsMultiStatementTransaction())
{
ereport(ERROR, (errmsg("worker_partition_query_result can only be used in a "

19
src/backend/distributed/metadata/dependency.c
View File

@ -17,6 +17,7 @@
#include "access/htup_details.h"
#include "access/skey.h"
#include "access/sysattr.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/indexing.h"
#include "catalog/pg_class.h"
@ -379,9 +380,19 @@ DependencyDefinitionFromPgShDepend(ObjectAddress target)
/*
* Scan pg_shdepend for dbid = $1 AND classid = $2 AND objid = $3 using
* pg_shdepend_depender_index
*
* where $1 is decided as follows:
* - shared dependencies $1 = InvalidOid
* - other dependencies $1 = MyDatabaseId
* This is consistent with postgres' static classIdGetDbId function
*/
Oid dbid = InvalidOid;
if (!IsSharedRelation(target.classId))
{
dbid = MyDatabaseId;
}
ScanKeyInit(&key[0], Anum_pg_shdepend_dbid, BTEqualStrategyNumber, F_OIDEQ,
MyDatabaseId);
ObjectIdGetDatum(dbid));
ScanKeyInit(&key[1], Anum_pg_shdepend_classid, BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(target.classId));
ScanKeyInit(&key[2], Anum_pg_shdepend_objid, BTEqualStrategyNumber, F_OIDEQ,
@ -570,6 +581,12 @@ SupportedDependencyByCitus(const ObjectAddress *address)
return true;
}
case OCLASS_DATABASE:
{
/* only to propagate its owner */
return true;
}
case OCLASS_ROLE:
{
/*

31
src/backend/distributed/metadata/metadata_cache.c
View File

@ -2641,6 +2641,8 @@ SecondaryNodeRoleId(void)
Datum
citus_dist_partition_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
TriggerData *triggerData = (TriggerData *) fcinfo->context;
Oid oldLogicalRelationId = InvalidOid;
Oid newLogicalRelationId = InvalidOid;
@ -2651,8 +2653,6 @@ citus_dist_partition_cache_invalidate(PG_FUNCTION_ARGS)
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
if (RelationGetRelid(triggerData->tg_relation) != DistPartitionRelationId())
{
ereport(ERROR, (errcode(ERRCODE_E_R_I_E_TRIGGER_PROTOCOL_VIOLATED),
@ -2718,6 +2718,8 @@ master_dist_partition_cache_invalidate(PG_FUNCTION_ARGS)
Datum
citus_dist_shard_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
TriggerData *triggerData = (TriggerData *) fcinfo->context;
Oid oldLogicalRelationId = InvalidOid;
Oid newLogicalRelationId = InvalidOid;
@ -2728,8 +2730,6 @@ citus_dist_shard_cache_invalidate(PG_FUNCTION_ARGS)
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
if (RelationGetRelid(triggerData->tg_relation) != DistShardRelationId())
{
ereport(ERROR, (errcode(ERRCODE_E_R_I_E_TRIGGER_PROTOCOL_VIOLATED),
@ -2795,6 +2795,8 @@ master_dist_shard_cache_invalidate(PG_FUNCTION_ARGS)
Datum
citus_dist_placement_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
TriggerData *triggerData = (TriggerData *) fcinfo->context;
Oid oldShardId = InvalidOid;
Oid newShardId = InvalidOid;
@ -2805,8 +2807,6 @@ citus_dist_placement_cache_invalidate(PG_FUNCTION_ARGS)
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
/*
* Before 7.0-2 this trigger is on pg_dist_shard_placement,
* ignore trigger in this scenario.
@ -2884,14 +2884,14 @@ master_dist_placement_cache_invalidate(PG_FUNCTION_ARGS)
Datum
citus_dist_node_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
if (!CALLED_AS_TRIGGER(fcinfo))
{
ereport(ERROR, (errcode(ERRCODE_E_R_I_E_TRIGGER_PROTOCOL_VIOLATED),
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
CitusInvalidateRelcacheByRelid(DistNodeRelationId());
PG_RETURN_DATUM(PointerGetDatum(NULL));
@ -2919,14 +2919,14 @@ master_dist_node_cache_invalidate(PG_FUNCTION_ARGS)
Datum
citus_conninfo_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
if (!CALLED_AS_TRIGGER(fcinfo))
{
ereport(ERROR, (errcode(ERRCODE_E_R_I_E_TRIGGER_PROTOCOL_VIOLATED),
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
/* no-op in community edition */
PG_RETURN_DATUM(PointerGetDatum(NULL));
@ -2954,14 +2954,14 @@ master_dist_authinfo_cache_invalidate(PG_FUNCTION_ARGS)
Datum
citus_dist_local_group_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
if (!CALLED_AS_TRIGGER(fcinfo))
{
ereport(ERROR, (errcode(ERRCODE_E_R_I_E_TRIGGER_PROTOCOL_VIOLATED),
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
CitusInvalidateRelcacheByRelid(DistLocalGroupIdRelationId());
PG_RETURN_DATUM(PointerGetDatum(NULL));
@ -2989,14 +2989,14 @@ master_dist_local_group_cache_invalidate(PG_FUNCTION_ARGS)
Datum
citus_dist_object_cache_invalidate(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
if (!CALLED_AS_TRIGGER(fcinfo))
{
ereport(ERROR, (errcode(ERRCODE_E_R_I_E_TRIGGER_PROTOCOL_VIOLATED),
errmsg("must be called as trigger")));
}
CheckCitusVersion(ERROR);
CitusInvalidateRelcacheByRelid(DistObjectRelationId());
PG_RETURN_DATUM(PointerGetDatum(NULL));
@ -3344,8 +3344,7 @@ GetLocalGroupId(void)
return LocalGroupId;
}
Oid localGroupTableOid = get_relname_relid("pg_dist_local_group",
PG_CATALOG_NAMESPACE);
Oid localGroupTableOid = DistLocalGroupIdRelationId();
if (localGroupTableOid == InvalidOid)
{
return 0;

244
src/backend/distributed/metadata/metadata_sync.c
View File

@ -25,11 +25,13 @@
#include "access/xact.h"
#include "catalog/dependency.h"
#include "catalog/indexing.h"
#include "catalog/pg_attrdef.h"
#include "catalog/pg_depend.h"
#include "catalog/pg_foreign_server.h"
#include "catalog/pg_namespace.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
#include "commands/sequence.h"
#include "distributed/citus_ruleutils.h"
#include "distributed/commands.h"
#include "distributed/deparser.h"
@ -46,10 +48,13 @@
#include "distributed/pg_dist_node.h"
#include "distributed/remote_commands.h"
#include "distributed/worker_manager.h"
#include "distributed/worker_protocol.h"
#include "distributed/worker_transaction.h"
#include "distributed/version_compat.h"
#include "executor/spi.h"
#include "foreign/foreign.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
#include "nodes/pg_list.h"
#include "pgstat.h"
#include "postmaster/bgworker.h"
@ -67,7 +72,6 @@ static List * GetDistributedTableDDLEvents(Oid relationId);
static char * LocalGroupIdUpdateCommand(int32 groupId);
static void UpdateDistNodeBoolAttr(const char *nodeName, int32 nodePort,
int attrNum, bool value);
static List * SequenceDDLCommandsForTable(Oid relationId);
static List * SequenceDependencyCommandList(Oid relationId);
static char * TruncateTriggerCreateCommand(Oid relationId);
static char * SchemaOwnerName(Oid objectId);
@ -103,6 +107,8 @@ static bool got_SIGALRM = false;
Datum
start_metadata_sync_to_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeName = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
@ -126,10 +132,10 @@ StartMetadataSyncToNode(const char *nodeNameString, int32 nodePort)
/* fail if metadata synchronization doesn't succeed */
bool raiseInterrupts = true;
CheckCitusVersion(ERROR);
EnsureCoordinator();
EnsureSuperUser();
EnsureModificationsCanRun();
CheckCitusVersion(ERROR);
PreventInTransactionBlock(true, "start_metadata_sync_to_node");
@ -185,14 +191,14 @@ StartMetadataSyncToNode(const char *nodeNameString, int32 nodePort)
Datum
stop_metadata_sync_to_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
EnsureSuperUser();
text *nodeName = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
char *nodeNameString = text_to_cstring(nodeName);
EnsureCoordinator();
EnsureSuperUser();
CheckCitusVersion(ERROR);
LockRelationOid(DistNodeRelationId(), ExclusiveLock);
WorkerNode *workerNode = FindWorkerNode(nodeNameString, nodePort);
@ -372,11 +378,9 @@ MetadataCreateCommands(void)
continue;
}
List *workerSequenceDDLCommands = SequenceDDLCommandsForTable(relationId);
List *ddlCommandList = GetFullTableCreationCommands(relationId,
includeSequenceDefaults);
char *tableOwnerResetCommand = TableOwnerResetCommand(relationId);
List *sequenceDependencyCommandList = SequenceDependencyCommandList(relationId);
/*
* Tables might have dependencies on different objects, since we create shards for
@ -386,6 +390,7 @@ MetadataCreateCommands(void)
ObjectAddressSet(tableAddress, RelationRelationId, relationId);
EnsureDependenciesExistOnAllNodes(&tableAddress);
List *workerSequenceDDLCommands = SequenceDDLCommandsForTable(relationId);
metadataSnapshotCommandList = list_concat(metadataSnapshotCommandList,
workerSequenceDDLCommands);
@ -400,6 +405,9 @@ MetadataCreateCommands(void)
metadataSnapshotCommandList = lappend(metadataSnapshotCommandList,
tableOwnerResetCommand);
List *sequenceDependencyCommandList = SequenceDependencyCommandList(
relationId);
metadataSnapshotCommandList = list_concat(metadataSnapshotCommandList,
sequenceDependencyCommandList);
}
@ -507,7 +515,8 @@ GetDistributedTableDDLEvents(Oid relationId)
}
/* command to associate sequences with table */
List *sequenceDependencyCommandList = SequenceDependencyCommandList(relationId);
List *sequenceDependencyCommandList = SequenceDependencyCommandList(
relationId);
commandList = list_concat(commandList, sequenceDependencyCommandList);
}
@ -1040,21 +1049,58 @@ List *
SequenceDDLCommandsForTable(Oid relationId)
{
List *sequenceDDLList = NIL;
List *ownedSequences = GetSequencesOwnedByRelation(relationId);
List *attnumList = NIL;
List *dependentSequenceList = NIL;
GetDependentSequencesWithRelation(relationId, &attnumList, &dependentSequenceList, 0);
char *ownerName = TableOwner(relationId);
Oid sequenceOid = InvalidOid;
foreach_oid(sequenceOid, ownedSequences)
ListCell *attnumCell = NULL;
ListCell *dependentSequenceCell = NULL;
forboth(attnumCell, attnumList, dependentSequenceCell, dependentSequenceList)
{
AttrNumber attnum = lfirst_int(attnumCell);
Oid sequenceOid = lfirst_oid(dependentSequenceCell);
char *sequenceDef = pg_get_sequencedef_string(sequenceOid);
char *escapedSequenceDef = quote_literal_cstr(sequenceDef);
StringInfo wrappedSequenceDef = makeStringInfo();
StringInfo sequenceGrantStmt = makeStringInfo();
char *sequenceName = generate_qualified_relation_name(sequenceOid);
Form_pg_sequence sequenceData = pg_get_sequencedef(sequenceOid);
Oid sequenceTypeOid = sequenceData->seqtypid;
Oid sequenceTypeOid = GetAttributeTypeOid(relationId, attnum);
char *typeName = format_type_be(sequenceTypeOid);
/* get sequence address */
ObjectAddress sequenceAddress = { 0 };
ObjectAddressSet(sequenceAddress, RelationRelationId, sequenceOid);
EnsureDependenciesExistOnAllNodes(&sequenceAddress);
/*
* Alter the sequence's data type in the coordinator if needed.
* A sequence's type is bigint by default and it doesn't change even if
* it's used in an int column. However, when distributing the sequence,
* we don't allow incompatible min/max ranges between the coordinator and
* workers, so we determine the sequence type here based on its current usage
* and propagate that same type to the workers as well.
* TODO: move this command to the part where the sequence is
* used in a distributed table: both in create_distributed_table
* and ALTER TABLE commands that include a sequence default
*/
Oid currentSequenceTypeOid = sequenceData->seqtypid;
if (currentSequenceTypeOid != sequenceTypeOid)
{
AlterSeqStmt *alterSequenceStatement = makeNode(AlterSeqStmt);
char *seqNamespace = get_namespace_name(get_rel_namespace(sequenceOid));
char *seqName = get_rel_name(sequenceOid);
alterSequenceStatement->sequence = makeRangeVar(seqNamespace, seqName, -1);
Node *asTypeNode = (Node *) makeTypeNameFromOid(sequenceTypeOid, -1);
SetDefElemArg(alterSequenceStatement, "as", asTypeNode);
ParseState *pstate = make_parsestate(NULL);
AlterSequence(pstate, alterSequenceStatement);
}
/* create schema if needed */
appendStringInfo(wrappedSequenceDef,
WORKER_APPLY_SEQUENCE_COMMAND,
@ -1067,12 +1113,184 @@ SequenceDDLCommandsForTable(Oid relationId)
sequenceDDLList = lappend(sequenceDDLList, wrappedSequenceDef->data);
sequenceDDLList = lappend(sequenceDDLList, sequenceGrantStmt->data);
MarkObjectDistributed(&sequenceAddress);
}
return sequenceDDLList;
}
/*
* GetAttributeTypeOid returns the OID of the type of the attribute of
* provided relationId that has the provided attnum
*/
Oid
GetAttributeTypeOid(Oid relationId, AttrNumber attnum)
{
Oid resultOid = InvalidOid;
ScanKeyData key[2];
/* Grab an appropriate lock on the pg_attribute relation */
Relation attrel = table_open(AttributeRelationId, AccessShareLock);
/* Use the index to scan only system attributes of the target relation */
ScanKeyInit(&key[0],
Anum_pg_attribute_attrelid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(relationId));
ScanKeyInit(&key[1],
Anum_pg_attribute_attnum,
BTLessEqualStrategyNumber, F_INT2LE,
Int16GetDatum(attnum));
SysScanDesc scan = systable_beginscan(attrel, AttributeRelidNumIndexId, true, NULL, 2,
key);
HeapTuple attributeTuple;
while (HeapTupleIsValid(attributeTuple = systable_getnext(scan)))
{
Form_pg_attribute att = (Form_pg_attribute) GETSTRUCT(attributeTuple);
resultOid = att->atttypid;
}
systable_endscan(scan);
table_close(attrel, AccessShareLock);
return resultOid;
}
/*
* GetDependentSequencesWithRelation appends the attnum and id of sequences that
* have direct (owned sequences) or indirect dependency with the given relationId,
* to the lists passed as NIL initially.
* For both cases, we use the intermediate AttrDefault object from pg_depend.
* If attnum is specified, we only return the sequences related to that
* attribute of the relationId.
*/
void
GetDependentSequencesWithRelation(Oid relationId, List **attnumList,
List **dependentSequenceList, AttrNumber attnum)
{
Assert(*attnumList == NIL && *dependentSequenceList == NIL);
List *attrdefResult = NIL;
List *attrdefAttnumResult = NIL;
ScanKeyData key[3];
HeapTuple tup;
Relation depRel = table_open(DependRelationId, AccessShareLock);
ScanKeyInit(&key[0],
Anum_pg_depend_refclassid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationRelationId));
ScanKeyInit(&key[1],
Anum_pg_depend_refobjid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(relationId));
if (attnum)
{
ScanKeyInit(&key[2],
Anum_pg_depend_refobjsubid,
BTEqualStrategyNumber, F_INT4EQ,
Int32GetDatum(attnum));
}
SysScanDesc scan = systable_beginscan(depRel, DependReferenceIndexId, true,
NULL, attnum ? 3 : 2, key);
while (HeapTupleIsValid(tup = systable_getnext(scan)))
{
Form_pg_depend deprec = (Form_pg_depend) GETSTRUCT(tup);
if (deprec->classid == AttrDefaultRelationId &&
deprec->objsubid == 0 &&
deprec->refobjsubid != 0 &&
deprec->deptype == DEPENDENCY_AUTO)
{
attrdefResult = lappend_oid(attrdefResult, deprec->objid);
attrdefAttnumResult = lappend_int(attrdefAttnumResult, deprec->refobjsubid);
}
}
systable_endscan(scan);
table_close(depRel, AccessShareLock);
ListCell *attrdefOidCell = NULL;
ListCell *attrdefAttnumCell = NULL;
forboth(attrdefOidCell, attrdefResult, attrdefAttnumCell, attrdefAttnumResult)
{
Oid attrdefOid = lfirst_oid(attrdefOidCell);
AttrNumber attrdefAttnum = lfirst_int(attrdefAttnumCell);
List *sequencesFromAttrDef = GetSequencesFromAttrDef(attrdefOid);
/* to simplify and eliminate cases like "DEFAULT nextval('..') - nextval('..')" */
if (list_length(sequencesFromAttrDef) > 1)
{
ereport(ERROR, (errmsg("More than one sequence in a column default"
" is not supported for distribution")));
}
if (list_length(sequencesFromAttrDef) == 1)
{
*dependentSequenceList = list_concat(*dependentSequenceList,
sequencesFromAttrDef);
*attnumList = lappend_int(*attnumList, attrdefAttnum);
}
}
}
/*
* GetSequencesFromAttrDef returns a list of sequence OIDs that have
* dependency with the given attrdefOid in pg_depend
*/
List *
GetSequencesFromAttrDef(Oid attrdefOid)
{
List *sequencesResult = NIL;
ScanKeyData key[2];
HeapTuple tup;
Relation depRel = table_open(DependRelationId, AccessShareLock);
ScanKeyInit(&key[0],
Anum_pg_depend_classid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(AttrDefaultRelationId));
ScanKeyInit(&key[1],
Anum_pg_depend_objid,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(attrdefOid));
SysScanDesc scan = systable_beginscan(depRel, DependDependerIndexId, true,
NULL, 2, key);
while (HeapTupleIsValid(tup = systable_getnext(scan)))
{
Form_pg_depend deprec = (Form_pg_depend) GETSTRUCT(tup);
if (deprec->refclassid == RelationRelationId &&
deprec->deptype == DEPENDENCY_NORMAL &&
get_rel_relkind(deprec->refobjid) == RELKIND_SEQUENCE)
{
sequencesResult = lappend_oid(sequencesResult, deprec->refobjid);
}
}
systable_endscan(scan);
table_close(depRel, AccessShareLock);
return sequencesResult;
}
/*
* SequenceDependencyCommandList generates commands to record the dependency
* of sequences on tables on the worker. This dependency does not exist by

44
src/backend/distributed/metadata/metadata_utility.c
View File

@ -270,13 +270,13 @@ citus_shard_sizes(PG_FUNCTION_ARGS)
Datum
citus_total_relation_size(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
bool failOnError = PG_GETARG_BOOL(1);
SizeQueryType sizeQueryType = TOTAL_RELATION_SIZE;
CheckCitusVersion(ERROR);
if (CStoreTable(relationId))
{
sizeQueryType = CSTORE_TABLE_SIZE;
@ -301,12 +301,12 @@ citus_total_relation_size(PG_FUNCTION_ARGS)
Datum
citus_table_size(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
bool failOnError = true;
SizeQueryType sizeQueryType = TABLE_SIZE;
CheckCitusVersion(ERROR);
if (CStoreTable(relationId))
{
sizeQueryType = CSTORE_TABLE_SIZE;
@ -331,12 +331,12 @@ citus_table_size(PG_FUNCTION_ARGS)
Datum
citus_relation_size(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
bool failOnError = true;
SizeQueryType sizeQueryType = RELATION_SIZE;
CheckCitusVersion(ERROR);
if (CStoreTable(relationId))
{
sizeQueryType = CSTORE_TABLE_SIZE;
@ -644,7 +644,19 @@ DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
StringInfo tableSizeStringInfo = (StringInfo) linitial(sizeList);
char *tableSizeString = tableSizeStringInfo->data;
if (strlen(tableSizeString) > 0)
{
*tableSize = SafeStringToUint64(tableSizeString);
}
else
{
/*
* This means the shard is moved or dropped while citus_total_relation_size is
* being executed. For this case we get an empty string as table size.
* We can take that as zero to prevent any unnecessary errors.
*/
*tableSize = 0;
}
PQclear(result);
ClearResults(connection, failOnError);
@ -1288,6 +1300,26 @@ ShardLength(uint64 shardId)
}
/*
* NodeGroupHasLivePlacements returns true if there is any placement
* on the given node group which is not a SHARD_STATE_TO_DELETE placement.
*/
bool
NodeGroupHasLivePlacements(int32 groupId)
{
List *shardPlacements = AllShardPlacementsOnNodeGroup(groupId);
GroupShardPlacement *placement = NULL;
foreach_ptr(placement, shardPlacements)
{
if (placement->shardState != SHARD_STATE_TO_DELETE)
{
return true;
}
}
return false;
}
/*
* NodeGroupHasShardPlacements returns whether any active shards are placed on the group
*/

64
src/backend/distributed/metadata/node_metadata.c
View File

@ -112,7 +112,7 @@ static bool UnsetMetadataSyncedForAll(void);
static void ErrorIfCoordinatorMetadataSetFalse(WorkerNode *workerNode, Datum value,
char *field);
static WorkerNode * SetShouldHaveShards(WorkerNode *workerNode, bool shouldHaveShards);
static void RemoveOldShardPlacementForNodeGroup(int groupId);
/* declarations for dynamic loading */
PG_FUNCTION_INFO_V1(citus_set_coordinator_host);
@ -161,6 +161,8 @@ DefaultNodeMetadata()
Datum
citus_set_coordinator_host(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeName = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
char *nodeNameString = text_to_cstring(nodeName);
@ -173,8 +175,6 @@ citus_set_coordinator_host(PG_FUNCTION_ARGS)
Name nodeClusterName = PG_GETARG_NAME(3);
nodeMetadata.nodeCluster = NameStr(*nodeClusterName);
CheckCitusVersion(ERROR);
/* prevent concurrent modification */
LockRelationOid(DistNodeRelationId(), RowShareLock);
@ -219,6 +219,8 @@ citus_set_coordinator_host(PG_FUNCTION_ARGS)
Datum
citus_add_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeName = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
char *nodeNameString = text_to_cstring(nodeName);
@ -227,8 +229,6 @@ citus_add_node(PG_FUNCTION_ARGS)
bool nodeAlreadyExists = false;
nodeMetadata.groupId = PG_GETARG_INT32(2);
CheckCitusVersion(ERROR);
/*
* During tests this function is called before nodeRole and nodeCluster have been
* created.
@ -288,6 +288,8 @@ master_add_node(PG_FUNCTION_ARGS)
Datum
citus_add_inactive_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeName = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
char *nodeNameString = text_to_cstring(nodeName);
@ -299,8 +301,6 @@ citus_add_inactive_node(PG_FUNCTION_ARGS)
nodeMetadata.nodeRole = PG_GETARG_OID(3);
nodeMetadata.nodeCluster = NameStr(*nodeClusterName);
CheckCitusVersion(ERROR);
if (nodeMetadata.groupId == COORDINATOR_GROUP_ID)
{
ereport(ERROR, (errmsg("coordinator node cannot be added as inactive node")));
@ -331,6 +331,8 @@ master_add_inactive_node(PG_FUNCTION_ARGS)
Datum
citus_add_secondary_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeName = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
char *nodeNameString = text_to_cstring(nodeName);
@ -348,8 +350,6 @@ citus_add_secondary_node(PG_FUNCTION_ARGS)
nodeMetadata.nodeRole = SecondaryNodeRoleId();
nodeMetadata.isActive = true;
CheckCitusVersion(ERROR);
int nodeId = AddNodeMetadata(nodeNameString, nodePort, &nodeMetadata,
&nodeAlreadyExists);
TransactionModifiedNodeMetadata = true;
@ -380,11 +380,11 @@ master_add_secondary_node(PG_FUNCTION_ARGS)
Datum
citus_remove_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeNameText = PG_GETARG_TEXT_P(0);
int32 nodePort = PG_GETARG_INT32(1);
CheckCitusVersion(ERROR);
RemoveNodeFromCluster(text_to_cstring(nodeNameText), nodePort);
TransactionModifiedNodeMetadata = true;
@ -631,7 +631,6 @@ static WorkerNode *
ModifiableWorkerNode(const char *nodeName, int32 nodePort)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
/* take an exclusive lock on pg_dist_node to serialize pg_dist_node changes */
@ -843,6 +842,8 @@ ActivateNode(char *nodeName, int nodePort)
Datum
citus_update_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
int32 nodeId = PG_GETARG_INT32(0);
text *newNodeName = PG_GETARG_TEXT_P(1);
@ -864,8 +865,6 @@ citus_update_node(PG_FUNCTION_ARGS)
List *placementList = NIL;
BackgroundWorkerHandle *handle = NULL;
CheckCitusVersion(ERROR);
WorkerNode *workerNodeWithSameAddress = FindWorkerNodeAnyCluster(newNodeNameString,
newNodePort);
if (workerNodeWithSameAddress != NULL)
@ -1077,10 +1076,10 @@ UpdateNodeLocation(int32 nodeId, char *newNodeName, int32 newNodePort)
Datum
get_shard_id_for_distribution_column(PG_FUNCTION_ARGS)
{
ShardInterval *shardInterval = NULL;
CheckCitusVersion(ERROR);
ShardInterval *shardInterval = NULL;
/*
* To have optional parameter as NULL, we defined this UDF as not strict, therefore
* we need to check all parameters for NULL values.
@ -1291,11 +1290,9 @@ RemoveNodeFromCluster(char *nodeName, int32 nodePort)
*/
DeleteAllReferenceTablePlacementsFromNodeGroup(workerNode->groupId);
}
bool onlyConsiderActivePlacements = false;
if (NodeGroupHasShardPlacements(workerNode->groupId,
onlyConsiderActivePlacements))
if (NodeGroupHasLivePlacements(workerNode->groupId))
{
if (ClusterHasReferenceTable())
if (ActivePrimaryNodeCount() == 1 && ClusterHasReferenceTable())
{
ereport(ERROR, (errmsg(
"cannot remove the last worker node because there are reference "
@ -1320,6 +1317,8 @@ RemoveNodeFromCluster(char *nodeName, int32 nodePort)
DeleteNodeRow(workerNode->workerName, nodePort);
RemoveOldShardPlacementForNodeGroup(workerNode->groupId);
char *nodeDeleteCommand = NodeDeleteCommand(workerNode->nodeId);
/* make sure we don't have any lingering session lifespan connections */
@ -1329,6 +1328,29 @@ RemoveNodeFromCluster(char *nodeName, int32 nodePort)
}
/*
* RemoveOldShardPlacementForNodeGroup removes all old shard placements
* for the given node group from pg_dist_placement.
*/
static void
RemoveOldShardPlacementForNodeGroup(int groupId)
{
/*
* Prevent concurrent deferred drop
*/
LockPlacementCleanup();
List *shardPlacementsOnNode = AllShardPlacementsOnNodeGroup(groupId);
GroupShardPlacement *placement = NULL;
foreach_ptr(placement, shardPlacementsOnNode)
{
if (placement->shardState == SHARD_STATE_TO_DELETE)
{
DeleteShardPlacementRow(placement->placementId);
}
}
}
/*
* CanRemoveReferenceTablePlacements returns true if active primary
* node count is more than 1, which means that even if we remove a node
@ -1818,7 +1840,7 @@ InsertPlaceholderCoordinatorRecord(void)
bool nodeAlreadyExists = false;
/* as long as there is a single node, localhost should be ok */
AddNodeMetadata("localhost", PostPortNumber, &nodeMetadata, &nodeAlreadyExists);
AddNodeMetadata(LocalHostName, PostPortNumber, &nodeMetadata, &nodeAlreadyExists);
}

4
src/backend/distributed/operations/citus_create_restore_point.c
View File

@ -49,12 +49,12 @@ PG_FUNCTION_INFO_V1(citus_create_restore_point);
Datum
citus_create_restore_point(PG_FUNCTION_ARGS)
{
text *restoreNameText = PG_GETARG_TEXT_P(0);
CheckCitusVersion(ERROR);
EnsureSuperUser();
EnsureCoordinator();
text *restoreNameText = PG_GETARG_TEXT_P(0);
if (RecoveryInProgress())
{
ereport(ERROR,

4
src/backend/distributed/operations/citus_tools.c
View File

@ -71,14 +71,14 @@ static Tuplestorestate * CreateTupleStore(TupleDesc tupleDescriptor,
Datum
master_run_on_worker(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
bool parallelExecution = false;
StringInfo *nodeNameArray = NULL;
int *nodePortArray = NULL;
StringInfo *commandStringArray = NULL;
CheckCitusVersion(ERROR);
/* check to see if caller supports us returning a tuplestore */
if (!rsinfo || !(rsinfo->allowedModes & SFRM_Materialize))
{

6
src/backend/distributed/operations/create_shards.c
View File

@ -64,6 +64,9 @@ PG_FUNCTION_INFO_V1(master_create_worker_shards);
Datum
master_create_worker_shards(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
text *tableNameText = PG_GETARG_TEXT_P(0);
int32 shardCount = PG_GETARG_INT32(1);
int32 replicationFactor = PG_GETARG_INT32(2);
@ -74,9 +77,6 @@ master_create_worker_shards(PG_FUNCTION_ARGS)
/* do not add any data */
bool useExclusiveConnections = false;
EnsureCoordinator();
CheckCitusVersion(ERROR);
/*
* distributed tables might have dependencies on different objects, since we create
* shards for a distributed table via multiple sessions these objects will be created

12
src/backend/distributed/operations/delete_protocol.c
View File

@ -109,6 +109,9 @@ PG_FUNCTION_INFO_V1(master_drop_sequences);
Datum
master_apply_delete_command(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
text *queryText = PG_GETARG_TEXT_P(0);
char *queryString = text_to_cstring(queryText);
List *deletableShardIntervalList = NIL;
@ -116,9 +119,6 @@ master_apply_delete_command(PG_FUNCTION_ARGS)
RawStmt *rawStmt = (RawStmt *) ParseTreeRawStmt(queryString);
Node *queryTreeNode = rawStmt->stmt;
EnsureCoordinator();
CheckCitusVersion(ERROR);
if (!IsA(queryTreeNode, DeleteStmt))
{
ereport(ERROR, (errmsg("query \"%s\" is not a delete statement",
@ -208,6 +208,8 @@ master_apply_delete_command(PG_FUNCTION_ARGS)
Datum
citus_drop_all_shards(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *schemaNameText = PG_GETARG_TEXT_P(1);
text *relationNameText = PG_GETARG_TEXT_P(2);
@ -215,8 +217,6 @@ citus_drop_all_shards(PG_FUNCTION_ARGS)
char *schemaName = text_to_cstring(schemaNameText);
char *relationName = text_to_cstring(relationNameText);
CheckCitusVersion(ERROR);
/*
* The SQL_DROP trigger calls this function even for tables that are
* not distributed. In that case, silently ignore and return -1.
@ -325,7 +325,7 @@ DropShards(Oid relationId, char *schemaName, char *relationName,
*/
if (MultiShardCommitProtocol == COMMIT_PROTOCOL_2PC)
{
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
}
List *dropTaskList = DropTaskList(relationId, schemaName, relationName,

4
src/backend/distributed/operations/modify_multiple_shards.c
View File

@ -70,13 +70,13 @@ PG_FUNCTION_INFO_V1(master_modify_multiple_shards);
Datum
master_modify_multiple_shards(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *queryText = PG_GETARG_TEXT_P(0);
char *queryString = text_to_cstring(queryText);
RawStmt *rawStmt = (RawStmt *) ParseTreeRawStmt(queryString);
Node *queryTreeNode = rawStmt->stmt;
CheckCitusVersion(ERROR);
if (!IsA(queryTreeNode, DeleteStmt) && !IsA(queryTreeNode, UpdateStmt))
{
ereport(ERROR, (errmsg("query \"%s\" is not a delete or update "

143
src/backend/distributed/operations/node_protocol.c
View File

@ -74,10 +74,13 @@ int ShardPlacementPolicy = SHARD_PLACEMENT_ROUND_ROBIN;
int NextShardId = 0;
int NextPlacementId = 0;
static List * GetTableReplicaIdentityCommand(Oid relationId);
static void GatherIndexAndConstraintDefinitionListExcludingReplicaIdentity(Form_pg_index
indexForm,
List **
indexDDLEventList,
int indexFlags);
static Datum WorkerNodeGetDatum(WorkerNode *workerNode, TupleDesc tupleDescriptor);
static void GatherIndexAndConstraintDefinitionList(Form_pg_index indexForm,
List **indexDDLEventList);
/* exports for SQL callable functions */
PG_FUNCTION_INFO_V1(master_get_table_metadata);
@ -100,6 +103,8 @@ PG_FUNCTION_INFO_V1(master_stage_shard_placement_row);
Datum
master_get_table_metadata(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *relationName = PG_GETARG_TEXT_P(0);
Oid relationId = ResolveRelationId(relationName, false);
@ -109,8 +114,6 @@ master_get_table_metadata(PG_FUNCTION_ARGS)
Datum values[TABLE_METADATA_FIELDS];
bool isNulls[TABLE_METADATA_FIELDS];
CheckCitusVersion(ERROR);
/* find partition tuple for partitioned relation */
CitusTableCacheEntry *partitionEntry = GetCitusTableCacheEntry(relationId);
@ -198,11 +201,11 @@ CStoreTable(Oid relationId)
Datum
master_get_table_ddl_events(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
FuncCallContext *functionContext = NULL;
ListCell *tableDDLEventCell = NULL;
CheckCitusVersion(ERROR);
/*
* On the very first call to this function, we first use the given relation
* name to get to the relation. We then recreate the list of DDL statements
@ -273,8 +276,8 @@ master_get_table_ddl_events(PG_FUNCTION_ARGS)
Datum
master_get_new_shardid(PG_FUNCTION_ARGS)
{
EnsureCoordinator();
CheckCitusVersion(ERROR);
EnsureCoordinator();
uint64 shardId = GetNextShardId();
Datum shardIdDatum = Int64GetDatum(shardId);
@ -343,8 +346,8 @@ GetNextShardId()
Datum
master_get_new_placementid(PG_FUNCTION_ARGS)
{
EnsureCoordinator();
CheckCitusVersion(ERROR);
EnsureCoordinator();
uint64 placementId = GetNextPlacementId();
Datum placementIdDatum = Int64GetDatum(placementId);
@ -450,11 +453,11 @@ master_stage_shard_placement_row(PG_FUNCTION_ARGS)
Datum
citus_get_active_worker_nodes(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
FuncCallContext *functionContext = NULL;
uint32 workerNodeCount = 0;
CheckCitusVersion(ERROR);
if (SRF_IS_FIRSTCALL())
{
/* create a function context for cross-call persistence */
@ -549,7 +552,7 @@ GetFullTableCreationCommands(Oid relationId, bool includeSequenceDefaults)
tableDDLEventList = list_concat(tableDDLEventList, preLoadCreationCommandList);
List *postLoadCreationCommandList =
GetPostLoadTableCreationCommands(relationId, true);
GetPostLoadTableCreationCommands(relationId, true, true);
tableDDLEventList = list_concat(tableDDLEventList, postLoadCreationCommandList);
@ -562,19 +565,43 @@ GetFullTableCreationCommands(Oid relationId, bool includeSequenceDefaults)
* of DDL commands that should be applied after loading the data.
*/
List *
GetPostLoadTableCreationCommands(Oid relationId, bool includeIndexes)
GetPostLoadTableCreationCommands(Oid relationId, bool includeIndexes,
bool includeReplicaIdentity)
{
List *tableDDLEventList = NIL;
if (includeIndexes)
/*
* Include all the commands (e.g., create index, set index clustered
* and set index statistics) regarding the indexes. Note that
* running all these commands in parallel might fail as the
* latter two depends on the first one. So, the caller should
* execute the commands sequentially.
*/
int indexFlags = INCLUDE_INDEX_ALL_STATEMENTS;
if (includeIndexes && includeReplicaIdentity)
{
List *indexAndConstraintCommandList =
GetTableIndexAndConstraintCommands(relationId);
GetTableIndexAndConstraintCommands(relationId, indexFlags);
tableDDLEventList = list_concat(tableDDLEventList, indexAndConstraintCommandList);
}
else if (includeIndexes && !includeReplicaIdentity)
{
/*
* Do not include the indexes/constraints that backs
* replica identity, if any.
*/
List *indexAndConstraintCommandList =
GetTableIndexAndConstraintCommandsExcludingReplicaIdentity(relationId,
indexFlags);
tableDDLEventList = list_concat(tableDDLEventList, indexAndConstraintCommandList);
}
if (includeReplicaIdentity)
{
List *replicaIdentityEvents = GetTableReplicaIdentityCommand(relationId);
tableDDLEventList = list_concat(tableDDLEventList, replicaIdentityEvents);
}
List *triggerCommands = GetExplicitTriggerCommandList(relationId);
tableDDLEventList = list_concat(tableDDLEventList, triggerCommands);
@ -590,7 +617,7 @@ GetPostLoadTableCreationCommands(Oid relationId, bool includeIndexes)
* GetTableReplicaIdentityCommand returns the list of DDL commands to
* (re)define the replica identity choice for a given table.
*/
static List *
List *
GetTableReplicaIdentityCommand(Oid relationId)
{
List *replicaIdentityCreateCommandList = NIL;
@ -694,18 +721,82 @@ GetPreLoadTableCreationCommands(Oid relationId, bool includeSequenceDefaults,
* (re)create indexes and constraints for a given table.
*/
List *
GetTableIndexAndConstraintCommands(Oid relationId)
GetTableIndexAndConstraintCommands(Oid relationId, int indexFlags)
{
return ExecuteFunctionOnEachTableIndex(relationId,
GatherIndexAndConstraintDefinitionList);
GatherIndexAndConstraintDefinitionList,
indexFlags);
}
/*
* GetTableIndexAndConstraintCommands returns the list of DDL commands to
* (re)create indexes and constraints for a given table.
*/
List *
GetTableIndexAndConstraintCommandsExcludingReplicaIdentity(Oid relationId, int indexFlags)
{
return ExecuteFunctionOnEachTableIndex(relationId,
GatherIndexAndConstraintDefinitionListExcludingReplicaIdentity,
indexFlags);
}
/*
* GatherIndexAndConstraintDefinitionListExcludingReplicaIdentity is a wrapper around
* GatherIndexAndConstraintDefinitionList(), which only excludes the indexes or
* constraints that back the replica identity.
*/
static void
GatherIndexAndConstraintDefinitionListExcludingReplicaIdentity(Form_pg_index indexForm,
List **indexDDLEventList,
int indexFlags)
{
Oid relationId = indexForm->indrelid;
Relation relation = table_open(relationId, AccessShareLock);
Oid replicaIdentityIndex = GetRelationIdentityOrPK(relation);
if (replicaIdentityIndex == indexForm->indexrelid)
{
/* this index is backing the replica identity, so skip */
table_close(relation, NoLock);
return;
}
GatherIndexAndConstraintDefinitionList(indexForm, indexDDLEventList, indexFlags);
table_close(relation, NoLock);
}
/*
* Get replica identity index or if it is not defined a primary key.
*
* If neither is defined, returns InvalidOid.
*
* Inspired from postgres/src/backend/replication/logical/worker.c
*/
Oid
GetRelationIdentityOrPK(Relation rel)
{
Oid idxoid = RelationGetReplicaIndex(rel);
if (!OidIsValid(idxoid))
{
idxoid = RelationGetPrimaryKeyIndex(rel);
}
return idxoid;
}
/*
* GatherIndexAndConstraintDefinitionList adds the DDL command for the given index.
*/
static void
GatherIndexAndConstraintDefinitionList(Form_pg_index indexForm, List **indexDDLEventList)
void
GatherIndexAndConstraintDefinitionList(Form_pg_index indexForm, List **indexDDLEventList,
int indexFlags)
{
Oid indexId = indexForm->indexrelid;
char *statementDef = NULL;
@ -726,11 +817,15 @@ GatherIndexAndConstraintDefinitionList(Form_pg_index indexForm, List **indexDDLE
}
/* append found constraint or index definition to the list */
if (indexFlags & INCLUDE_CREATE_INDEX_STATEMENTS)
{
*indexDDLEventList = lappend(*indexDDLEventList, makeTableDDLCommandString(
statementDef));
}
/* if table is clustered on this index, append definition to the list */
if (indexForm->indisclustered)
if ((indexFlags & INCLUDE_INDEX_CLUSTERED_STATEMENTS) &&
indexForm->indisclustered)
{
char *clusteredDef = pg_get_indexclusterdef_string(indexId);
Assert(clusteredDef != NULL);
@ -740,8 +835,12 @@ GatherIndexAndConstraintDefinitionList(Form_pg_index indexForm, List **indexDDLE
}
/* we need alter index commands for altered targets on expression indexes */
if (indexFlags & INCLUDE_INDEX_STATISTICS_STATEMENTTS)
{
List *alterIndexStatisticsCommands = GetAlterIndexStatisticsCommands(indexId);
*indexDDLEventList = list_concat(*indexDDLEventList, alterIndexStatisticsCommands);
*indexDDLEventList = list_concat(*indexDDLEventList,
alterIndexStatisticsCommands);
}
}

4
src/backend/distributed/operations/partitioning.c
View File

@ -35,10 +35,10 @@ PG_FUNCTION_INFO_V1(time_partition_range);
Datum
time_partition_range(PG_FUNCTION_ARGS)
{
Oid relationId = PG_GETARG_OID(0);
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
/* create tuple descriptor for return value */
TupleDesc metadataDescriptor = NULL;
TypeFuncClass resultTypeClass = get_call_result_type(fcinfo, NULL,

166
src/backend/distributed/operations/repair_shards.c
View File

@ -63,7 +63,10 @@ static void ReplicateColocatedShardPlacement(int64 shardId, char *sourceNodeName
char shardReplicationMode);
static void CopyShardTables(List *shardIntervalList, char *sourceNodeName,
int32 sourceNodePort, char *targetNodeName,
int32 targetNodePort);
int32 targetNodePort, bool useLogicalReplication);
static void CopyShardTablesViaBlockWrites(List *shardIntervalList, char *sourceNodeName,
int32 sourceNodePort,
char *targetNodeName, int32 targetNodePort);
static List * CopyPartitionShardsCommandList(ShardInterval *shardInterval,
const char *sourceNodeName,
int32 sourceNodePort);
@ -93,6 +96,7 @@ static void EnsureEnoughDiskSpaceForShardMove(List *colocatedShardList,
char *targetNodeName, uint32
targetNodePort);
/* declarations for dynamic loading */
PG_FUNCTION_INFO_V1(citus_copy_shard_placement);
PG_FUNCTION_INFO_V1(master_copy_shard_placement);
@ -118,6 +122,9 @@ bool CheckAvailableSpaceBeforeMove = true;
Datum
citus_copy_shard_placement(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
int64 shardId = PG_GETARG_INT64(0);
text *sourceNodeNameText = PG_GETARG_TEXT_P(1);
int32 sourceNodePort = PG_GETARG_INT32(2);
@ -129,9 +136,6 @@ citus_copy_shard_placement(PG_FUNCTION_ARGS)
char *sourceNodeName = text_to_cstring(sourceNodeNameText);
char *targetNodeName = text_to_cstring(targetNodeNameText);
EnsureCoordinator();
CheckCitusVersion(ERROR);
char shardReplicationMode = LookupShardTransferMode(shardReplicationModeOid);
if (shardReplicationMode == TRANSFER_MODE_FORCE_LOGICAL)
{
@ -279,6 +283,9 @@ CheckSpaceConstraints(MultiConnection *connection, uint64 colocationSizeInBytes)
Datum
citus_move_shard_placement(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
int64 shardId = PG_GETARG_INT64(0);
char *sourceNodeName = text_to_cstring(PG_GETARG_TEXT_P(1));
int32 sourceNodePort = PG_GETARG_INT32(2);
@ -290,12 +297,9 @@ citus_move_shard_placement(PG_FUNCTION_ARGS)
ListCell *colocatedTableCell = NULL;
ListCell *colocatedShardCell = NULL;
CheckCitusVersion(ERROR);
EnsureCoordinator();
Oid relationId = RelationIdForShard(shardId);
ErrorIfMoveCitusLocalTable(relationId);
ErrorIfTargetNodeIsNotSafeToMove(targetNodeName, targetNodePort);
ShardInterval *shardInterval = LoadShardInterval(shardId);
Oid distributedTableId = shardInterval->relationId;
@ -359,8 +363,9 @@ citus_move_shard_placement(PG_FUNCTION_ARGS)
* CopyColocatedShardPlacement function copies given shard with its co-located
* shards.
*/
bool useLogicalReplication = false;
CopyShardTables(colocatedShardList, sourceNodeName, sourceNodePort, targetNodeName,
targetNodePort);
targetNodePort, useLogicalReplication);
ShardInterval *colocatedShard = NULL;
foreach_ptr(colocatedShard, colocatedShardList)
@ -417,6 +422,51 @@ EnsureEnoughDiskSpaceForShardMove(List *colocatedShardList,
}
/*
* ErrorIfTargetNodeIsNotSafeToMove throws error if the target node is not
* eligible for moving shards.
*/
void
ErrorIfTargetNodeIsNotSafeToMove(const char *targetNodeName, int targetNodePort)
{
WorkerNode *workerNode = FindWorkerNode(targetNodeName, targetNodePort);
if (workerNode == NULL)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Moving shards to a non-existing node is not supported"),
errhint(
"Add the target node via SELECT citus_add_node('%s', %d);",
targetNodeName, targetNodePort)));
}
if (!workerNode->isActive)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Moving shards to a non-active node is not supported"),
errhint(
"Activate the target node via SELECT citus_activate_node('%s', %d);",
targetNodeName, targetNodePort)));
}
if (!workerNode->shouldHaveShards)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Moving shards to a node that shouldn't have a shard is "
"not supported"),
errhint("Allow shards on the target node via "
"SELECT * FROM citus_set_node_property('%s', %d, 'shouldhaveshards', true);",
targetNodeName, targetNodePort)));
}
if (!NodeIsPrimary(workerNode))
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("Moving shards to a secondary (e.g., replica) node is "
"not supported")));
}
}
/*
* master_move_shard_placement is a wrapper around citus_move_shard_placement.
*/
@ -741,8 +791,9 @@ ReplicateColocatedShardPlacement(int64 shardId, char *sourceNodeName,
EnsureReferenceTablesExistOnAllNodesExtended(shardReplicationMode);
}
bool useLogicalReplication = false;
CopyShardTables(colocatedShardList, sourceNodeName, sourceNodePort,
targetNodeName, targetNodePort);
targetNodeName, targetNodePort, useLogicalReplication);
/*
* Finally insert the placements to pg_dist_placement and sync it to the
@ -820,32 +871,51 @@ EnsureTableListSuitableForReplication(List *tableIdList)
/*
* CopyColocatedShardPlacement copies a shard along with its co-located shards
* from a source node to target node. It does not make any checks about state
* of the shards. It is caller's responsibility to make those checks if they are
* necessary.
* CopyShardTables copies a shard along with its co-located shards from a source
* node to target node. It does not make any checks about state of the shards.
* It is caller's responsibility to make those checks if they are necessary.
*/
static void
CopyShardTables(List *shardIntervalList, char *sourceNodeName, int32 sourceNodePort,
char *targetNodeName, int32 targetNodePort)
char *targetNodeName, int32 targetNodePort, bool useLogicalReplication)
{
ShardInterval *shardInterval = NULL;
if (list_length(shardIntervalList) < 1)
{
return;
}
if (useLogicalReplication)
{
/* only supported in Citus enterprise */
}
else
{
CopyShardTablesViaBlockWrites(shardIntervalList, sourceNodeName, sourceNodePort,
targetNodeName, targetNodePort);
}
}
/*
* CopyShardTablesViaBlockWrites copies a shard along with its co-located shards
* from a source node to target node via COPY command. While the command is in
* progress, the modifications on the source node is blocked.
*/
static void
CopyShardTablesViaBlockWrites(List *shardIntervalList, char *sourceNodeName,
int32 sourceNodePort, char *targetNodeName,
int32 targetNodePort)
{
MemoryContext localContext = AllocSetContextCreate(CurrentMemoryContext,
"CopyShardTables",
"CopyShardTablesViaBlockWrites",
ALLOCSET_DEFAULT_SIZES);
MemoryContext oldContext = MemoryContextSwitchTo(localContext);
/* iterate through the colocated shards and copy each */
ShardInterval *shardInterval = NULL;
foreach_ptr(shardInterval, shardIntervalList)
{
bool includeDataCopy = true;
if (PartitionedTable(shardInterval->relationId))
{
/* partitioned tables contain no data */
includeDataCopy = false;
}
bool includeDataCopy = !PartitionedTable(shardInterval->relationId);
List *ddlCommandList = CopyShardCommandList(shardInterval, sourceNodeName,
sourceNodePort, includeDataCopy);
@ -853,10 +923,9 @@ CopyShardTables(List *shardIntervalList, char *sourceNodeName, int32 sourceNodeP
SendCommandListToWorkerInSingleTransaction(targetNodeName, targetNodePort,
tableOwner, ddlCommandList);
}
MemoryContextReset(localContext);
}
/*
* Once all shards are created, we can recreate relationships between shards.
@ -868,15 +937,14 @@ CopyShardTables(List *shardIntervalList, char *sourceNodeName, int32 sourceNodeP
{
List *shardForeignConstraintCommandList = NIL;
List *referenceTableForeignConstraintList = NIL;
char *tableOwner = TableOwner(shardInterval->relationId);
List *commandList = NIL;
CopyShardForeignConstraintCommandListGrouped(shardInterval,
&shardForeignConstraintCommandList,
&referenceTableForeignConstraintList);
List *commandList = list_concat(shardForeignConstraintCommandList,
referenceTableForeignConstraintList);
commandList = list_concat(commandList, shardForeignConstraintCommandList);
commandList = list_concat(commandList, referenceTableForeignConstraintList);
if (PartitionTable(shardInterval->relationId))
{
@ -886,8 +954,10 @@ CopyShardTables(List *shardIntervalList, char *sourceNodeName, int32 sourceNodeP
commandList = lappend(commandList, attachPartitionCommand);
}
char *tableOwner = TableOwner(shardInterval->relationId);
SendCommandListToWorkerInSingleTransaction(targetNodeName, targetNodePort,
tableOwner, commandList);
MemoryContextReset(localContext);
}
@ -990,11 +1060,41 @@ EnsureShardCanBeCopied(int64 shardId, const char *sourceNodeName, int32 sourceNo
targetNodeName,
targetNodePort);
if (targetPlacement != NULL)
{
if (targetPlacement->shardState == SHARD_STATE_TO_DELETE)
{
/*
* Trigger deletion of orphaned shards and hope that this removes
* the shard.
*/
DropOrphanedShardsInSeparateTransaction();
shardPlacementList = ShardPlacementList(shardId);
targetPlacement = SearchShardPlacementInList(shardPlacementList,
targetNodeName,
targetNodePort);
/*
* If it still doesn't remove the shard, then we error.
*/
if (targetPlacement != NULL)
{
ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("shard " INT64_FORMAT " already exists in the target node",
errmsg(
"shard " INT64_FORMAT
" still exists on the target node as an orphaned shard",
shardId),
errdetail(
"The existing shard is orphaned, but could not be deleted because there are still active queries on it")));
}
}
else
{
ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg(
"shard " INT64_FORMAT " already exists in the target node",
shardId)));
}
}
}
@ -1079,7 +1179,9 @@ CopyShardCommandList(ShardInterval *shardInterval, const char *sourceNodeName,
copyShardDataCommand->data);
}
List *indexCommandList = GetPostLoadTableCreationCommands(relationId, true);
bool includeReplicaIdentity = true;
List *indexCommandList =
GetPostLoadTableCreationCommands(relationId, true, includeReplicaIdentity);
indexCommandList = WorkerApplyShardDDLCommandList(indexCommandList, shardId);
copyShardToNodeCommandsList = list_concat(copyShardToNodeCommandsList,

138
src/backend/distributed/operations/shard_cleaner.c
View File

@ -12,59 +12,109 @@
#include "postgres.h"
#include "access/xact.h"
#include "postmaster/postmaster.h"
#include "distributed/coordinator_protocol.h"
#include "distributed/metadata_cache.h"
#include "distributed/shard_cleaner.h"
#include "distributed/shard_rebalancer.h"
#include "distributed/remote_commands.h"
#include "distributed/resource_lock.h"
#include "distributed/worker_transaction.h"
/* declarations for dynamic loading */
PG_FUNCTION_INFO_V1(master_defer_delete_shards);
PG_FUNCTION_INFO_V1(citus_cleanup_orphaned_shards);
PG_FUNCTION_INFO_V1(isolation_cleanup_orphaned_shards);
static bool TryDropShard(GroupShardPlacement *placement);
static bool TryLockRelationAndPlacementCleanup(Oid relationId, LOCKMODE lockmode);
/*
* master_defer_delete_shards implements a user-facing UDF to deleter orphaned shards that
* are still haning around in the system. These shards are orphaned by previous actions
* that were not directly able to delete the placements eg. shard moving or dropping of a
* distributed table while one of the data nodes was not online.
* citus_cleanup_orphaned_shards implements a user-facing UDF to delete
* orphaned shards that are still haning around in the system. These shards are
* orphaned by previous actions that were not directly able to delete the
* placements eg. shard moving or dropping of a distributed table while one of
* the data nodes was not online.
*
* This function iterates through placements where shardstate is SHARD_STATE_TO_DELETE
* (shardstate = 4), drops the corresponding tables from the node and removes the
* placement information from the catalog.
* This function iterates through placements where shardstate is
* SHARD_STATE_TO_DELETE (shardstate = 4), drops the corresponding tables from
* the node and removes the placement information from the catalog.
*
* The function takes no arguments and runs cluster wide
* The function takes no arguments and runs cluster wide. It cannot be run in a
* transaction, because holding the locks it takes for a long time is not good.
* While the locks are held, it is impossible for the background daemon to
* cleanup orphaned shards.
*/
Datum
master_defer_delete_shards(PG_FUNCTION_ARGS)
citus_cleanup_orphaned_shards(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
PreventInTransactionBlock(true, "citus_cleanup_orphaned_shards");
bool waitForCleanupLock = true;
int droppedShardCount = DropMarkedShards(waitForCleanupLock);
bool waitForLocks = true;
int droppedShardCount = DropOrphanedShards(waitForLocks);
if (droppedShardCount > 0)
{
ereport(NOTICE, (errmsg("cleaned up %d orphaned shards", droppedShardCount)));
}
PG_RETURN_INT32(droppedShardCount);
PG_RETURN_VOID();
}
/*
* TryDropMarkedShards is a wrapper around DropMarkedShards that catches
* isolation_cleanup_orphaned_shards implements a test UDF that's the same as
* citus_cleanup_orphaned_shards. The only difference is that this command can
* be run in transactions, this is to test
*/
Datum
isolation_cleanup_orphaned_shards(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
bool waitForLocks = true;
int droppedShardCount = DropOrphanedShards(waitForLocks);
if (droppedShardCount > 0)
{
ereport(NOTICE, (errmsg("cleaned up %d orphaned shards", droppedShardCount)));
}
PG_RETURN_VOID();
}
/*
* DropOrphanedShardsInSeparateTransaction cleans up orphaned shards by
* connecting to localhost. This is done, so that the locks that
* DropOrphanedShards takes are only held for a short time.
*/
void
DropOrphanedShardsInSeparateTransaction(void)
{
ExecuteCriticalCommandInSeparateTransaction("CALL citus_cleanup_orphaned_shards()");
}
/*
* TryDropOrphanedShards is a wrapper around DropOrphanedShards that catches
* any errors to make it safe to use in the maintenance daemon.
*
* If dropping any of the shards failed this function returns -1, otherwise it
* returns the number of dropped shards.
*/
int
TryDropMarkedShards(bool waitForCleanupLock)
TryDropOrphanedShards(bool waitForLocks)
{
int droppedShardCount = 0;
MemoryContext savedContext = CurrentMemoryContext;
PG_TRY();
{
droppedShardCount = DropMarkedShards(waitForCleanupLock);
droppedShardCount = DropOrphanedShards(waitForLocks);
}
PG_CATCH();
{
@ -83,7 +133,7 @@ TryDropMarkedShards(bool waitForCleanupLock)
/*
* DropMarkedShards removes shards that were marked SHARD_STATE_TO_DELETE before.
* DropOrphanedShards removes shards that were marked SHARD_STATE_TO_DELETE before.
*
* It does so by trying to take an exclusive lock on the shard and its
* colocated placements before removing. If the lock cannot be obtained it
@ -91,33 +141,46 @@ TryDropMarkedShards(bool waitForCleanupLock)
* will be removed at a later time when there are no locks held anymore on
* those placements.
*
* If waitForLocks is false, then if we cannot take a lock on pg_dist_placement
* we continue without waiting.
*
* Before doing any of this it will take an exclusive PlacementCleanup lock.
* This is to ensure that this function is not being run concurrently.
* Otherwise really bad race conditions are possible, such as removing all
* placements of a shard. waitForCleanupLock indicates if this function should
* wait for this lock or error out.
* placements of a shard. waitForLocks indicates if this function should
* wait for this lock or not.
*
*/
int
DropMarkedShards(bool waitForCleanupLock)
DropOrphanedShards(bool waitForLocks)
{
int removedShardCount = 0;
ListCell *shardPlacementCell = NULL;
/*
* We should try to take the highest lock that we take
* later in this function for pg_dist_placement. We take RowExclusiveLock
* in DeleteShardPlacementRow.
*/
LOCKMODE lockmode = RowExclusiveLock;
if (!IsCoordinator())
{
return 0;
}
if (waitForCleanupLock)
if (waitForLocks)
{
LockPlacementCleanup();
}
else if (!TryLockPlacementCleanup())
else
{
Oid distPlacementId = DistPlacementRelationId();
if (!TryLockRelationAndPlacementCleanup(distPlacementId, lockmode))
{
ereport(WARNING, (errmsg("could not acquire lock to cleanup placements")));
return 0;
}
}
int failedShardDropCount = 0;
List *shardPlacementList = AllShardPlacementsWithShardPlacementState(
@ -145,7 +208,7 @@ DropMarkedShards(bool waitForCleanupLock)
if (failedShardDropCount > 0)
{
ereport(WARNING, (errmsg("Failed to drop %d old shards out of %d",
ereport(WARNING, (errmsg("Failed to drop %d orphaned shards out of %d",
failedShardDropCount, list_length(shardPlacementList))));
}
@ -153,10 +216,33 @@ DropMarkedShards(bool waitForCleanupLock)
}
/*
* TryLockRelationAndCleanup tries to lock the given relation
* and the placement cleanup. If it cannot, it returns false.
*
*/
static bool
TryLockRelationAndPlacementCleanup(Oid relationId, LOCKMODE lockmode)
{
if (!ConditionalLockRelationOid(relationId, lockmode))
{
ereport(DEBUG1, (errmsg(
"could not acquire shard lock to cleanup placements")));
return false;
}
if (!TryLockPlacementCleanup())
{
ereport(DEBUG1, (errmsg("could not acquire lock to cleanup placements")));
return false;
}
return true;
}
/*
* TryDropShard tries to drop the given shard placement and returns
* true on success. On failure, this method swallows errors and emits them
* as WARNINGs.
* true on success.
*/
static bool
TryDropShard(GroupShardPlacement *placement)

741
src/backend/distributed/operations/shard_rebalancer.c
View File

File diff suppressed because it is too large Load Diff

4
src/backend/distributed/operations/split_shards.c
View File

@ -61,10 +61,10 @@ isolate_tenant_to_new_shard(PG_FUNCTION_ARGS)
Datum
worker_hash(PG_FUNCTION_ARGS)
{
Datum valueDatum = PG_GETARG_DATUM(0);
CheckCitusVersion(ERROR);
Datum valueDatum = PG_GETARG_DATUM(0);
/* figure out hash function from the data type */
Oid valueDataType = get_fn_expr_argtype(fcinfo->flinfo, 0);
TypeCacheEntry *typeEntry = lookup_type_cache(valueDataType,

20
src/backend/distributed/operations/stage_protocol.c
View File

@ -94,6 +94,8 @@ PG_FUNCTION_INFO_V1(citus_update_table_statistics);
Datum
master_create_empty_shard(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *relationNameText = PG_GETARG_TEXT_P(0);
char *relationName = text_to_cstring(relationNameText);
uint32 attemptableNodeCount = 0;
@ -108,8 +110,6 @@ master_create_empty_shard(PG_FUNCTION_ARGS)
Oid relationId = ResolveRelationId(relationNameText, false);
char relationKind = get_rel_relkind(relationId);
CheckCitusVersion(ERROR);
EnsureTablePermissions(relationId, ACL_INSERT);
CheckDistributedTable(relationId);
@ -171,10 +171,6 @@ master_create_empty_shard(PG_FUNCTION_ARGS)
"on local tables")));
}
char replicationModel = TableReplicationModel(relationId);
EnsureReplicationSettings(relationId, replicationModel);
/* generate new and unique shardId from sequence */
uint64 shardId = GetNextShardId();
@ -243,6 +239,8 @@ master_create_empty_shard(PG_FUNCTION_ARGS)
Datum
master_append_table_to_shard(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
uint64 shardId = PG_GETARG_INT64(0);
text *sourceTableNameText = PG_GETARG_TEXT_P(1);
text *sourceNodeNameText = PG_GETARG_TEXT_P(2);
@ -253,8 +251,6 @@ master_append_table_to_shard(PG_FUNCTION_ARGS)
float4 shardFillLevel = 0.0;
CheckCitusVersion(ERROR);
ShardInterval *shardInterval = LoadShardInterval(shardId);
Oid relationId = shardInterval->relationId;
@ -363,10 +359,10 @@ master_append_table_to_shard(PG_FUNCTION_ARGS)
Datum
citus_update_shard_statistics(PG_FUNCTION_ARGS)
{
int64 shardId = PG_GETARG_INT64(0);
CheckCitusVersion(ERROR);
int64 shardId = PG_GETARG_INT64(0);
uint64 shardSize = UpdateShardStatistics(shardId);
PG_RETURN_INT64(shardSize);
@ -380,10 +376,10 @@ citus_update_shard_statistics(PG_FUNCTION_ARGS)
Datum
citus_update_table_statistics(PG_FUNCTION_ARGS)
{
Oid distributedTableId = PG_GETARG_OID(0);
CheckCitusVersion(ERROR);
Oid distributedTableId = PG_GETARG_OID(0);
UpdateTableStatistics(distributedTableId);
PG_RETURN_VOID();

48
src/backend/distributed/planner/deparse_shard_query.c
View File

@ -59,6 +59,7 @@ RebuildQueryStrings(Job *workerJob)
Query *originalQuery = workerJob->jobQuery;
List *taskList = workerJob->taskList;
Task *task = NULL;
bool isSingleTask = list_length(taskList) == 1;
if (originalQuery->commandType == CMD_INSERT)
{
@ -74,7 +75,7 @@ RebuildQueryStrings(Job *workerJob)
* task, we scribble on the original query to avoid the copying
* overhead.
*/
if (list_length(taskList) > 1)
if (!isSingleTask)
{
query = copyObject(originalQuery);
}
@ -119,6 +120,19 @@ RebuildQueryStrings(Job *workerJob)
* deparse_shard_query when the string is needed
*/
task->anchorDistributedTableId = modifiedRelationRTE->relid;
/*
* For multi-row inserts, we modify the VALUES before storing the
* query in the task.
*/
RangeTblEntry *valuesRTE = ExtractDistributedInsertValuesRTE(query);
if (valuesRTE != NULL)
{
Assert(valuesRTE->rtekind == RTE_VALUES);
Assert(task->rowValuesLists != NULL);
valuesRTE->values_lists = task->rowValuesLists;
}
}
bool isQueryObjectOrText = GetTaskQueryType(task) == TASK_QUERY_TEXT ||
@ -180,39 +194,7 @@ AddInsertAliasIfNeeded(Query *query)
static void
UpdateTaskQueryString(Query *query, Task *task)
{
List *oldValuesLists = NIL;
RangeTblEntry *valuesRTE = NULL;
if (query->commandType == CMD_INSERT)
{
/* extract the VALUES from the INSERT */
valuesRTE = ExtractDistributedInsertValuesRTE(query);
if (valuesRTE != NULL)
{
Assert(valuesRTE->rtekind == RTE_VALUES);
Assert(task->rowValuesLists != NULL);
oldValuesLists = valuesRTE->values_lists;
valuesRTE->values_lists = task->rowValuesLists;
}
if (ShouldLazyDeparseQuery(task))
{
/*
* not all insert queries are copied before calling this
* function, so we do it here
*/
query = copyObject(query);
}
}
SetTaskQueryIfShouldLazyDeparse(task, query);
if (valuesRTE != NULL)
{
valuesRTE->values_lists = oldValuesLists;
}
}

10
src/backend/distributed/planner/local_distributed_join_planner.c
View File

@ -94,6 +94,7 @@
#include "distributed/multi_physical_planner.h"
#include "distributed/multi_server_executor.h"
#include "distributed/multi_router_planner.h"
#include "distributed/coordinator_protocol.h"
#include "distributed/query_colocation_checker.h"
#include "distributed/query_pushdown_planning.h"
#include "distributed/recursive_planning.h"
@ -179,7 +180,8 @@ static ConversionCandidates * CreateConversionCandidates(PlannerRestrictionConte
plannerRestrictionContext,
List *rangeTableList,
int resultRTEIdentity);
static void AppendUniqueIndexColumnsToList(Form_pg_index indexForm, List **uniqueIndexes);
static void AppendUniqueIndexColumnsToList(Form_pg_index indexForm, List **uniqueIndexes,
int flags);
static ConversionChoice GetConversionChoice(ConversionCandidates *
conversionCandidates,
PlannerRestrictionContext *
@ -403,7 +405,8 @@ HasConstantFilterOnUniqueColumn(RangeTblEntry *rangeTableEntry,
FetchEqualityAttrNumsForRTE((Node *) restrictClauseList);
List *uniqueIndexColumnsList = ExecuteFunctionOnEachTableIndex(rangeTableEntry->relid,
AppendUniqueIndexColumnsToList);
AppendUniqueIndexColumnsToList,
INCLUDE_INDEX_ALL_STATEMENTS);
IndexColumns *indexColumns = NULL;
foreach_ptr(indexColumns, uniqueIndexColumnsList)
{
@ -442,7 +445,8 @@ FirstIsSuperSetOfSecond(List *firstIntList, List *secondIntList)
* unique index.
*/
static void
AppendUniqueIndexColumnsToList(Form_pg_index indexForm, List **uniqueIndexGroups)
AppendUniqueIndexColumnsToList(Form_pg_index indexForm, List **uniqueIndexGroups,
int flags)
{
if (indexForm->indisunique || indexForm->indisprimary)
{

8
src/backend/distributed/planner/local_plan_cache.c
View File

@ -139,6 +139,14 @@ GetCachedLocalPlan(Task *task, DistributedPlan *distributedPlan)
bool
IsLocalPlanCachingSupported(Job *currentJob, DistributedPlan *originalDistributedPlan)
{
if (originalDistributedPlan->numberOfTimesExecuted < 1)
{
/*
* Only cache if a plan is being reused (via a prepared statement).
*/
return false;
}
if (!currentJob->deferredPruning)
{
/*

130
src/backend/distributed/planner/multi_logical_optimizer.c
View File

@ -318,10 +318,6 @@ static Node * WorkerLimitCount(Node *limitCount, Node *limitOffset, OrderByLimit
static List * WorkerSortClauseList(Node *limitCount,
List *groupClauseList, List *sortClauseList,
OrderByLimitReference orderByLimitReference);
static List * GenerateNewTargetEntriesForSortClauses(List *originalTargetList,
List *sortClauseList,
AttrNumber *targetProjectionNumber,
Index *nextSortGroupRefIndex);
static bool CanPushDownLimitApproximate(List *sortClauseList, List *targetList);
static bool HasOrderByAggregate(List *sortClauseList, List *targetList);
static bool HasOrderByNonCommutativeAggregate(List *sortClauseList, List *targetList);
@ -2701,38 +2697,6 @@ ProcessWindowFunctionsForWorkerQuery(List *windowClauseList,
return;
}
WindowClause *windowClause = NULL;
foreach_ptr(windowClause, windowClauseList)
{
List *partitionClauseTargetList =
GenerateNewTargetEntriesForSortClauses(originalTargetEntryList,
windowClause->partitionClause,
&(queryTargetList->
targetProjectionNumber),
queryWindowClause->
nextSortGroupRefIndex);
List *orderClauseTargetList =
GenerateNewTargetEntriesForSortClauses(originalTargetEntryList,
windowClause->orderClause,
&(queryTargetList->
targetProjectionNumber),
queryWindowClause->
nextSortGroupRefIndex);
/*
* Note that even Citus does push down the window clauses as-is, we may still need to
* add the generated entries to the target list. The reason is that the same aggregates
* might be referred from another target entry that is a bare aggregate (e.g., no window
* functions), which would have been mutated. For instance, when an average aggregate
* is mutated on the target list, the window function would refer to a sum aggregate,
* which is obviously wrong.
*/
queryTargetList->targetEntryList = list_concat(queryTargetList->targetEntryList,
partitionClauseTargetList);
queryTargetList->targetEntryList = list_concat(queryTargetList->targetEntryList,
orderClauseTargetList);
}
queryWindowClause->workerWindowClauseList = windowClauseList;
queryWindowClause->hasWindowFunctions = true;
}
@ -2798,19 +2762,6 @@ ProcessLimitOrderByForWorkerQuery(OrderByLimitReference orderByLimitReference,
groupClauseList,
sortClauseList,
orderByLimitReference);
/*
* TODO: Do we really need to add the target entries if we're not pushing
* down ORDER BY?
*/
List *newTargetEntryListForSortClauses =
GenerateNewTargetEntriesForSortClauses(originalTargetList,
queryOrderByLimit->workerSortClauseList,
&(queryTargetList->targetProjectionNumber),
queryOrderByLimit->nextSortGroupRefIndex);
queryTargetList->targetEntryList =
list_concat(queryTargetList->targetEntryList, newTargetEntryListForSortClauses);
}
@ -4795,87 +4746,6 @@ WorkerSortClauseList(Node *limitCount, List *groupClauseList, List *sortClauseLi
}
/*
* GenerateNewTargetEntriesForSortClauses goes over provided sort clause lists and
* creates new target entries if needed to make sure sort clauses has correct
* references. The function returns list of new target entries, caller is
* responsible to add those target entries to the end of worker target list.
*
* The function is required because we change the target entry if it contains an
* expression having an aggregate operation, or just the AVG aggregate.
* Afterwards any order by clause referring to original target entry starts
* to point to a wrong expression.
*
* Note the function modifies SortGroupClause items in sortClauseList,
* targetProjectionNumber, and nextSortGroupRefIndex.
*/
static List *
GenerateNewTargetEntriesForSortClauses(List *originalTargetList,
List *sortClauseList,
AttrNumber *targetProjectionNumber,
Index *nextSortGroupRefIndex)
{
List *createdTargetList = NIL;
SortGroupClause *sgClause = NULL;
foreach_ptr(sgClause, sortClauseList)
{
TargetEntry *targetEntry = get_sortgroupclause_tle(sgClause, originalTargetList);
Expr *targetExpr = targetEntry->expr;
bool containsAggregate = contain_aggs_of_level((Node *) targetExpr, 0);
bool createNewTargetEntry = false;
/* we are only interested in target entries containing aggregates */
if (!containsAggregate)
{
continue;
}
/*
* If the target expression is not an Aggref, it is either an expression
* on a single aggregate, or expression containing multiple aggregates.
* Worker query mutates these target entries to have a naked target entry
* per aggregate function. We want to use original target entries if this
* the case.
* If the original target expression is an avg aggref, we also want to use
* original target entry.
*/
if (!IsA(targetExpr, Aggref))
{
createNewTargetEntry = true;
}
else
{
Aggref *aggNode = (Aggref *) targetExpr;
AggregateType aggregateType = GetAggregateType(aggNode);
if (aggregateType == AGGREGATE_AVERAGE)
{
createNewTargetEntry = true;
}
}
if (createNewTargetEntry)
{
bool resJunk = true;
AttrNumber nextResNo = (*targetProjectionNumber);
Expr *newExpr = copyObject(targetExpr);
TargetEntry *newTargetEntry = makeTargetEntry(newExpr, nextResNo,
targetEntry->resname, resJunk);
newTargetEntry->ressortgroupref = *nextSortGroupRefIndex;
createdTargetList = lappend(createdTargetList, newTargetEntry);
sgClause->tleSortGroupRef = *nextSortGroupRefIndex;
(*nextSortGroupRefIndex)++;
(*targetProjectionNumber)++;
}
}
return createdTargetList;
}
/*
* CanPushDownLimitApproximate checks if we can push down the limit clause to
* the worker nodes, and get approximate and meaningful results. We can do this

2
src/backend/distributed/planner/multi_router_planner.c
View File

@ -2425,7 +2425,7 @@ CreateLocalDummyPlacement()
{
ShardPlacement *dummyPlacement = CitusMakeNode(ShardPlacement);
dummyPlacement->nodeId = LOCAL_NODE_ID;
dummyPlacement->nodeName = LOCAL_HOST_NAME;
dummyPlacement->nodeName = LocalHostName;
dummyPlacement->nodePort = PostPortNumber;
dummyPlacement->groupId = GetLocalGroupId();
return dummyPlacement;

4
src/backend/distributed/relay/relay_event_utility.c
View File

@ -901,13 +901,13 @@ AppendShardIdToName(char **name, uint64 shardId)
Datum
shard_name(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
int64 shardId = PG_GETARG_INT64(1);
char *qualifiedName = NULL;
CheckCitusVersion(ERROR);
if (shardId <= 0)
{
ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),

176
src/backend/distributed/shared_library_init.c
View File

@ -68,6 +68,7 @@
#include "distributed/time_constants.h"
#include "distributed/query_stats.h"
#include "distributed/remote_commands.h"
#include "distributed/shard_rebalancer.h"
#include "distributed/shared_library_init.h"
#include "distributed/statistics_collection.h"
#include "distributed/subplan_execution.h"
@ -98,6 +99,9 @@ PG_MODULE_MAGIC;
#define DUMMY_REAL_TIME_EXECUTOR_ENUM_VALUE 9999999
static char *CitusVersion = CITUS_VERSION;
/* deprecated GUC value that should not be used anywhere outside this file */
static int ReplicationModel = REPLICATION_MODEL_STREAMING;
void _PG_init(void);
void _PG_fini(void);
@ -114,6 +118,7 @@ static void RegisterCitusConfigVariables(void);
static bool ErrorIfNotASuitableDeadlockFactor(double *newval, void **extra,
GucSource source);
static bool WarnIfDeprecatedExecutorUsed(int *newval, void **extra, GucSource source);
static bool WarnIfReplicationModelIsSet(int *newval, void **extra, GucSource source);
static bool NoticeIfSubqueryPushdownEnabled(bool *newval, void **extra, GucSource source);
static bool NodeConninfoGucCheckHook(char **newval, void **extra, GucSource source);
static void NodeConninfoGucAssignHook(const char *newval, void *extra);
@ -575,6 +580,17 @@ RegisterCitusConfigVariables(void)
GUC_STANDARD,
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.check_available_space_before_move",
gettext_noop("When enabled will check free disk space before a shard move"),
gettext_noop(
"Free disk space will be checked when this setting is enabled before each shard move."),
&CheckAvailableSpaceBeforeMove,
true,
PGC_USERSET,
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomStringVariable(
"citus.cluster_name",
gettext_noop("Which cluster this node is a part of"),
@ -629,7 +645,9 @@ RegisterCitusConfigVariables(void)
DefineCustomBoolVariable(
"citus.defer_drop_after_shard_move",
gettext_noop("When enabled a shard move will mark old shards for deletion"),
gettext_noop("When enabled a shard move will mark the original shards "
"for deletion after a successful move, instead of deleting "
"them right away."),
gettext_noop("The deletion of a shard can sometimes run into a conflict with a "
"long running transactions on a the shard during the drop phase of "
"the shard move. This causes some moves to be rolled back after "
@ -639,7 +657,7 @@ RegisterCitusConfigVariables(void)
"citus.defer_shard_delete_interval to make sure defered deletions "
"will be executed"),
&DeferShardDeleteOnMove,
false,
true,
PGC_USERSET,
0,
NULL, NULL, NULL);
@ -654,11 +672,37 @@ RegisterCitusConfigVariables(void)
"the background worker moves on. When set to -1 this background "
"process is skipped."),
&DeferShardDeleteInterval,
-1, -1, 7 * 24 * 3600 * 1000,
15000, -1, 7 * 24 * 3600 * 1000,
PGC_SIGHUP,
GUC_UNIT_MS,
NULL, NULL, NULL);
DefineCustomRealVariable(
"citus.desired_percent_disk_available_after_move",
gettext_noop(
"Sets how many percentage of free disk space should be after a shard move"),
gettext_noop(
"This setting controls how much free space should be available after a shard move."
"If the free disk space will be lower than this parameter, then shard move will result in"
"an error."),
&DesiredPercentFreeAfterMove,
10.0, 0.0, 100.0,
PGC_SIGHUP,
GUC_STANDARD,
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.enable_manual_changes_to_shards",
gettext_noop("Enables dropping and truncating known shards."),
gettext_noop("Set to false by default. If set to true, enables "
"dropping and truncating shards on the coordinator "
"(or the workers with metadata)"),
&EnableManualChangesToShards,
false,
PGC_USERSET,
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomRealVariable(
"citus.distributed_deadlock_detection_factor",
gettext_noop("Sets the time to wait before checking for distributed "
@ -693,6 +737,17 @@ RegisterCitusConfigVariables(void)
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.enable_alter_database_owner",
gettext_noop("Enables propagating ALTER DATABASE ... OWNER TO ... statements to "
"workers"),
NULL,
&EnableAlterDatabaseOwner,
false,
PGC_USERSET,
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.enable_binary_protocol",
gettext_noop(
@ -940,30 +995,17 @@ RegisterCitusConfigVariables(void)
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.check_available_space_before_move",
gettext_noop("When enabled will check free disk space before a shard move"),
gettext_noop(
"Free disk space will be checked when this setting is enabled before each shard move."),
&CheckAvailableSpaceBeforeMove,
"citus.enable_cost_based_connection_establishment",
gettext_noop("When enabled the connection establishment times "
"and task execution times into account for deciding "
"whether or not to establish new connections."),
NULL,
&EnableCostBasedConnectionEstablishment,
true,
PGC_USERSET,
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomRealVariable(
"citus.desired_percent_disk_available_after_move",
gettext_noop(
"Sets how many percentage of free disk space should be after a shard move"),
gettext_noop(
"This setting controls how much free space should be available after a shard move."
"If the free disk space will be lower than this parameter, then shard move will result in"
"an error."),
&DesiredPercentFreeAfterMove,
10.0, 0.0, 100.0,
PGC_SIGHUP,
GUC_STANDARD,
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.explain_distributed_queries",
gettext_noop("Enables Explain for distributed queries."),
@ -1051,6 +1093,19 @@ RegisterCitusConfigVariables(void)
GUC_UNIT_BYTE | GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomStringVariable(
"citus.local_hostname",
gettext_noop("Sets the hostname when connecting back to itself."),
gettext_noop("For some operations nodes, mostly the coordinator, connect back to "
"itself. When configuring SSL certificates it sometimes is required "
"to use a specific hostname to match the CN of the certificate when "
"verify-full is used."),
&LocalHostName,
"localhost",
PGC_SUSET,
GUC_STANDARD,
NULL, NULL, NULL);
DefineCustomIntVariable(
"citus.local_shared_pool_size",
gettext_noop(
@ -1190,6 +1245,16 @@ RegisterCitusConfigVariables(void)
GUC_UNIT_KB | GUC_STANDARD,
NULL, NULL, NULL);
DefineCustomIntVariable(
"citus.max_rebalancer_logged_ignored_moves",
gettext_noop("Sets the maximum number of ignored moves the rebalance logs"),
NULL,
&MaxRebalancerLoggedIgnoredMoves,
5, -1, INT_MAX,
PGC_USERSET,
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomIntVariable(
"citus.max_shared_pool_size",
gettext_noop("Sets the maximum number of connections allowed per worker node "
@ -1364,6 +1429,21 @@ RegisterCitusConfigVariables(void)
GUC_UNIT_KB | GUC_STANDARD,
NULL, NULL, NULL);
DefineCustomBoolVariable(
"citus.prevent_incomplete_connection_establishment",
gettext_noop("When enabled, the executor waits until all the connections "
"are successfully established."),
gettext_noop("Under some load, the executor may decide to establish some "
"extra connections to further parallelize the execution. However,"
"before the connection establishment is done, the execution might "
"have already finished. When this GUC is set to true, the execution "
"waits for such connections to be established."),
&PreventIncompleteConnectionEstablishment,
true,
PGC_USERSET,
GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomEnumVariable(
"citus.propagate_set_commands",
gettext_noop("Sets which SET commands are propagated to workers."),
@ -1438,16 +1518,28 @@ RegisterCitusConfigVariables(void)
DefineCustomEnumVariable(
"citus.replication_model",
gettext_noop("Sets the replication model to be used for distributed tables."),
gettext_noop("Depending upon the execution environment, statement- or streaming-"
"based replication modes may be employed. Though most Citus deploy-"
"ments will simply use statement replication, hosted and MX-style"
"deployments should set this parameter to 'streaming'."),
gettext_noop("Deprecated. Please use citus.shard_replication_factor instead"),
gettext_noop(
"Shard replication model is determined by the shard replication factor."
"'statement' replication is used only when the replication factor is one."),
&ReplicationModel,
REPLICATION_MODEL_COORDINATOR,
REPLICATION_MODEL_STREAMING,
replication_model_options,
PGC_SUSET,
GUC_SUPERUSER_ONLY,
GUC_NO_SHOW_ALL,
WarnIfReplicationModelIsSet, NULL, NULL);
DefineCustomBoolVariable(
"citus.running_under_isolation_test",
gettext_noop(
"Only useful for testing purposes, when set to true, Citus does some "
"tricks to implement useful isolation tests with rebalancing. Should "
"never be set to true on production systems "),
gettext_noop("for details of the tricks implemented, refer to the source code"),
&RunningUnderIsolationTest,
false,
PGC_SUSET,
GUC_SUPERUSER_ONLY | GUC_NO_SHOW_ALL,
NULL, NULL, NULL);
DefineCustomBoolVariable(
@ -1741,6 +1833,32 @@ NoticeIfSubqueryPushdownEnabled(bool *newval, void **extra, GucSource source)
}
/*
* WarnIfReplicationModelIsSet prints a warning when a user sets
* citus.replication_model.
*/
static bool
WarnIfReplicationModelIsSet(int *newval, void **extra, GucSource source)
{
/* print a warning only when user sets the guc */
if (source == PGC_S_SESSION)
{
ereport(NOTICE, (errcode(ERRCODE_WARNING_DEPRECATED_FEATURE),
errmsg(
"Setting citus.replication_model has no effect. Please use "
"citus.shard_replication_factor instead."),
errdetail(
"Citus determines the replication model based on the "
"replication factor and the replication models of the colocated "
"shards. If a colocated table is present, the replication model "
"is inherited. Otherwise 'streaming' replication is preferred if "
"supported by the replication factor.")));
}
return true;
}
/*
* NodeConninfoGucCheckHook ensures conninfo settings are in the expected form
* and that the keywords of all non-null settings are on a allowlist devised to

42
src/backend/distributed/sql/citus--10.0-3--10.1-1.sql
View File

@ -1,9 +1,51 @@
-- citus--10.0-3--10.1-1
-- add the current database to the distributed objects if not already in there.
-- this is to reliably propagate some of the alter database commands that might be
-- supported.
INSERT INTO citus.pg_dist_object SELECT
'pg_catalog.pg_database'::regclass::oid AS oid,
(SELECT oid FROM pg_database WHERE datname = current_database()) as objid,
0 as objsubid
ON CONFLICT DO NOTHING;
#include "../../columnar/sql/columnar--10.0-3--10.1-1.sql"
#include "udfs/create_distributed_table/10.1-1.sql";
#include "udfs/worker_partitioned_relation_total_size/10.1-1.sql"
#include "udfs/worker_partitioned_relation_size/10.1-1.sql"
#include "udfs/worker_partitioned_table_size/10.1-1.sql"
#include "udfs/citus_prepare_pg_upgrade/10.1-1.sql"
#include "udfs/citus_finish_pg_upgrade/10.1-1.sql"
#include "udfs/citus_local_disk_space_stats/10.1-1.sql"
#include "udfs/get_rebalance_table_shards_plan/10.1-1.sql"
#include "udfs/citus_add_rebalance_strategy/10.1-1.sql"
ALTER TABLE pg_catalog.pg_dist_rebalance_strategy ADD COLUMN improvement_threshold float4 NOT NULL default 0;
UPDATE pg_catalog.pg_dist_rebalance_strategy SET improvement_threshold = 0.5 WHERE name = 'by_disk_size';
#include "udfs/get_rebalance_progress/10.1-1.sql"
-- use streaming replication when replication factor = 1
WITH replicated_shards AS (
SELECT shardid
FROM pg_dist_placement
WHERE shardstate = 1 OR shardstate = 3
GROUP BY shardid
HAVING count(*) <> 1 ),
replicated_relations AS (
SELECT DISTINCT logicalrelid
FROM pg_dist_shard
JOIN replicated_shards
USING (shardid)
)
UPDATE pg_dist_partition
SET repmodel = 's'
WHERE repmodel = 'c'
AND partmethod = 'h'
AND logicalrelid NOT IN (SELECT * FROM replicated_relations);
#include "udfs/citus_shards/10.1-1.sql"
DROP TRIGGER pg_dist_rebalance_strategy_enterprise_check_trigger ON pg_catalog.pg_dist_rebalance_strategy;
DROP FUNCTION citus_internal.pg_dist_rebalance_strategy_enterprise_check();
#include "udfs/citus_cleanup_orphaned_shards/10.1-1.sql"

4
src/backend/distributed/sql/citus--10.1-1--10.2-1.sql Normal file
View File

@ -0,0 +1,4 @@
-- citus--10.1-1--10.2-1
-- bump version to 10.2-1

19
src/backend/distributed/sql/citus--8.3-1--9.0-1.sql
View File

@ -83,24 +83,7 @@ DROP EXTENSION IF EXISTS shard_rebalancer;
#include "udfs/get_rebalance_table_shards_plan/9.0-1.sql"
#include "udfs/replicate_table_shards/9.0-1.sql"
#include "udfs/rebalance_table_shards/9.0-1.sql"
-- get_rebalance_progress returns the list of shard placement move operations along with
-- their progressions for ongoing rebalance operations.
--
CREATE OR REPLACE FUNCTION get_rebalance_progress()
RETURNS TABLE(sessionid integer,
table_name regclass,
shardid bigint,
shard_size bigint,
sourcename text,
sourceport int,
targetname text,
targetport int,
progress bigint)
AS 'MODULE_PATHNAME'
LANGUAGE C STRICT;
COMMENT ON FUNCTION get_rebalance_progress()
IS 'provides progress information about the ongoing rebalance operations';
#include "udfs/get_rebalance_progress/9.0-1.sql"
DROP FUNCTION master_add_node(text, integer, integer, noderole, name);
CREATE FUNCTION master_add_node(nodename text,

63
src/backend/distributed/sql/downgrades/citus--10.1-1--10.0-3.sql
View File

@ -1,5 +1,10 @@
-- citus--10.1-1--10.0-3
-- remove databases as distributed objects to prevent unknown object types being managed
-- on older versions.
DELETE FROM citus.pg_dist_object
WHERE classid = 'pg_catalog.pg_database'::regclass::oid;
#include "../../../columnar/sql/downgrades/columnar--10.1-1--10.0-3.sql"
DROP FUNCTION pg_catalog.create_distributed_table(regclass, text, citus.distribution_type, text, int);
@ -21,4 +26,62 @@ DROP FUNCTION pg_catalog.worker_partitioned_relation_size(regclass);
DROP FUNCTION pg_catalog.worker_partitioned_table_size(regclass);
DROP FUNCTION pg_catalog.citus_local_disk_space_stats();
#include "../udfs/citus_prepare_pg_upgrade/9.5-1.sql"
#include "../udfs/citus_finish_pg_upgrade/10.0-1.sql"
#include "../udfs/get_rebalance_table_shards_plan/9.2-1.sql"
-- the migration for citus_add_rebalance_strategy from 9.2-1 was the first one,
-- so it doesn't have a DROP. This is why we DROP manually here.
DROP FUNCTION pg_catalog.citus_add_rebalance_strategy;
#include "../udfs/citus_add_rebalance_strategy/9.2-1.sql"
ALTER TABLE pg_catalog.pg_dist_rebalance_strategy DROP COLUMN improvement_threshold;
-- the migration for get_rebalance_progress from 9.0-1 was the first one,
-- so it doesn't have a DROP. This is why we DROP manually here.
DROP FUNCTION pg_catalog.get_rebalance_progress;
#include "../udfs/get_rebalance_progress/9.0-1.sql"
CREATE OR REPLACE VIEW pg_catalog.citus_shards AS
WITH shard_sizes AS (SELECT * FROM pg_catalog.citus_shard_sizes())
SELECT
pg_dist_shard.logicalrelid AS table_name,
pg_dist_shard.shardid,
shard_name(pg_dist_shard.logicalrelid, pg_dist_shard.shardid) as shard_name,
CASE WHEN partkey IS NOT NULL THEN 'distributed' WHEN repmodel = 't' THEN 'reference' ELSE 'local' END AS citus_table_type,
colocationid AS colocation_id,
pg_dist_node.nodename,
pg_dist_node.nodeport,
(SELECT size FROM shard_sizes WHERE
shard_name(pg_dist_shard.logicalrelid, pg_dist_shard.shardid) = table_name
OR
'public.' || shard_name(pg_dist_shard.logicalrelid, pg_dist_shard.shardid) = table_name
LIMIT 1) as shard_size
FROM
pg_dist_shard
JOIN
pg_dist_placement
ON
pg_dist_shard.shardid = pg_dist_placement.shardid
JOIN
pg_dist_node
ON
pg_dist_placement.groupid = pg_dist_node.groupid
JOIN
pg_dist_partition
ON
pg_dist_partition.logicalrelid = pg_dist_shard.logicalrelid
ORDER BY
pg_dist_shard.logicalrelid::text, shardid
;
#include "../udfs/citus_finish_pg_upgrade/10.0-1.sql"
CREATE FUNCTION citus_internal.pg_dist_rebalance_strategy_enterprise_check()
RETURNS TRIGGER
LANGUAGE C
AS 'MODULE_PATHNAME';
CREATE TRIGGER pg_dist_rebalance_strategy_enterprise_check_trigger
BEFORE INSERT OR UPDATE OR DELETE OR TRUNCATE ON pg_dist_rebalance_strategy
FOR EACH STATEMENT EXECUTE FUNCTION citus_internal.pg_dist_rebalance_strategy_enterprise_check();
DROP PROCEDURE pg_catalog.citus_cleanup_orphaned_shards();

2
src/backend/distributed/sql/downgrades/citus--10.2-1--10.1-1.sql Normal file
View File

@ -0,0 +1,2 @@
-- citus--10.2-1--10.1-1
-- this is an empty downgrade path since citus--10.1-1--10.2-1.sql is empty for now

30
src/backend/distributed/sql/udfs/citus_add_rebalance_strategy/10.1-1.sql generated Normal file
View File

@ -0,0 +1,30 @@
DROP FUNCTION pg_catalog.citus_add_rebalance_strategy;
CREATE OR REPLACE FUNCTION pg_catalog.citus_add_rebalance_strategy(
name name,
shard_cost_function regproc,
node_capacity_function regproc,
shard_allowed_on_node_function regproc,
default_threshold float4,
minimum_threshold float4 DEFAULT 0,
improvement_threshold float4 DEFAULT 0
)
RETURNS VOID AS $$
INSERT INTO
pg_catalog.pg_dist_rebalance_strategy(
name,
shard_cost_function,
node_capacity_function,
shard_allowed_on_node_function,
default_threshold,
minimum_threshold
) VALUES (
name,
shard_cost_function,
node_capacity_function,
shard_allowed_on_node_function,
default_threshold,
minimum_threshold
);
$$ LANGUAGE sql;
COMMENT ON FUNCTION pg_catalog.citus_add_rebalance_strategy(name,regproc,regproc,regproc,float4, float4, float4)
IS 'adds a new rebalance strategy which can be used when rebalancing shards or draining nodes';

6
src/backend/distributed/sql/udfs/citus_add_rebalance_strategy/latest.sql
View File

@ -1,10 +1,12 @@
DROP FUNCTION pg_catalog.citus_add_rebalance_strategy;
CREATE OR REPLACE FUNCTION pg_catalog.citus_add_rebalance_strategy(
name name,
shard_cost_function regproc,
node_capacity_function regproc,
shard_allowed_on_node_function regproc,
default_threshold float4,
minimum_threshold float4 DEFAULT 0
minimum_threshold float4 DEFAULT 0,
improvement_threshold float4 DEFAULT 0
)
RETURNS VOID AS $$
INSERT INTO
@ -24,5 +26,5 @@ CREATE OR REPLACE FUNCTION pg_catalog.citus_add_rebalance_strategy(
minimum_threshold
);
$$ LANGUAGE sql;
COMMENT ON FUNCTION pg_catalog.citus_add_rebalance_strategy(name,regproc,regproc,regproc,float4, float4)
COMMENT ON FUNCTION pg_catalog.citus_add_rebalance_strategy(name,regproc,regproc,regproc,float4, float4, float4)
IS 'adds a new rebalance strategy which can be used when rebalancing shards or draining nodes';

5
src/backend/distributed/sql/udfs/citus_cleanup_orphaned_shards/10.1-1.sql generated Normal file
View File

@ -0,0 +1,5 @@
CREATE OR REPLACE PROCEDURE pg_catalog.citus_cleanup_orphaned_shards()
LANGUAGE C
AS 'citus', $$citus_cleanup_orphaned_shards$$;
COMMENT ON PROCEDURE pg_catalog.citus_cleanup_orphaned_shards()
IS 'cleanup orphaned shards';

5
src/backend/distributed/sql/udfs/citus_cleanup_orphaned_shards/latest.sql Normal file
View File

@ -0,0 +1,5 @@
CREATE OR REPLACE PROCEDURE pg_catalog.citus_cleanup_orphaned_shards()
LANGUAGE C
AS 'citus', $$citus_cleanup_orphaned_shards$$;
COMMENT ON PROCEDURE pg_catalog.citus_cleanup_orphaned_shards()
IS 'cleanup orphaned shards';

5
src/backend/distributed/sql/udfs/citus_finish_pg_upgrade/10.1-1.sql generated
View File

@ -23,7 +23,6 @@ BEGIN
INSERT INTO pg_catalog.pg_dist_authinfo SELECT * FROM public.pg_dist_authinfo;
INSERT INTO pg_catalog.pg_dist_poolinfo SELECT * FROM public.pg_dist_poolinfo;
ALTER TABLE pg_catalog.pg_dist_rebalance_strategy DISABLE TRIGGER pg_dist_rebalance_strategy_enterprise_check_trigger;
INSERT INTO pg_catalog.pg_dist_rebalance_strategy SELECT
name,
default_strategy,
@ -31,9 +30,9 @@ BEGIN
node_capacity_function::regprocedure::regproc,
shard_allowed_on_node_function::regprocedure::regproc,
default_threshold,
minimum_threshold
minimum_threshold,
improvement_threshold
FROM public.pg_dist_rebalance_strategy;
ALTER TABLE pg_catalog.pg_dist_rebalance_strategy ENABLE TRIGGER pg_dist_rebalance_strategy_enterprise_check_trigger;
--
-- drop backup tables

5
src/backend/distributed/sql/udfs/citus_finish_pg_upgrade/latest.sql
View File

@ -23,7 +23,6 @@ BEGIN
INSERT INTO pg_catalog.pg_dist_authinfo SELECT * FROM public.pg_dist_authinfo;
INSERT INTO pg_catalog.pg_dist_poolinfo SELECT * FROM public.pg_dist_poolinfo;
ALTER TABLE pg_catalog.pg_dist_rebalance_strategy DISABLE TRIGGER pg_dist_rebalance_strategy_enterprise_check_trigger;
INSERT INTO pg_catalog.pg_dist_rebalance_strategy SELECT
name,
default_strategy,
@ -31,9 +30,9 @@ BEGIN
node_capacity_function::regprocedure::regproc,
shard_allowed_on_node_function::regprocedure::regproc,
default_threshold,
minimum_threshold
minimum_threshold,
improvement_threshold
FROM public.pg_dist_rebalance_strategy;
ALTER TABLE pg_catalog.pg_dist_rebalance_strategy ENABLE TRIGGER pg_dist_rebalance_strategy_enterprise_check_trigger;
--
-- drop backup tables

54
src/backend/distributed/sql/udfs/citus_prepare_pg_upgrade/10.1-1.sql generated Normal file
View File

@ -0,0 +1,54 @@
CREATE OR REPLACE FUNCTION pg_catalog.citus_prepare_pg_upgrade()
RETURNS void
LANGUAGE plpgsql
SET search_path = pg_catalog
AS $cppu$
BEGIN
--
-- Drop existing backup tables
--
DROP TABLE IF EXISTS public.pg_dist_partition;
DROP TABLE IF EXISTS public.pg_dist_shard;
DROP TABLE IF EXISTS public.pg_dist_placement;
DROP TABLE IF EXISTS public.pg_dist_node_metadata;
DROP TABLE IF EXISTS public.pg_dist_node;
DROP TABLE IF EXISTS public.pg_dist_local_group;
DROP TABLE IF EXISTS public.pg_dist_transaction;
DROP TABLE IF EXISTS public.pg_dist_colocation;
DROP TABLE IF EXISTS public.pg_dist_authinfo;
DROP TABLE IF EXISTS public.pg_dist_poolinfo;
DROP TABLE IF EXISTS public.pg_dist_rebalance_strategy;
--
-- backup citus catalog tables
--
CREATE TABLE public.pg_dist_partition AS SELECT * FROM pg_catalog.pg_dist_partition;
CREATE TABLE public.pg_dist_shard AS SELECT * FROM pg_catalog.pg_dist_shard;
CREATE TABLE public.pg_dist_placement AS SELECT * FROM pg_catalog.pg_dist_placement;
CREATE TABLE public.pg_dist_node_metadata AS SELECT * FROM pg_catalog.pg_dist_node_metadata;
CREATE TABLE public.pg_dist_node AS SELECT * FROM pg_catalog.pg_dist_node;
CREATE TABLE public.pg_dist_local_group AS SELECT * FROM pg_catalog.pg_dist_local_group;
CREATE TABLE public.pg_dist_transaction AS SELECT * FROM pg_catalog.pg_dist_transaction;
CREATE TABLE public.pg_dist_colocation AS SELECT * FROM pg_catalog.pg_dist_colocation;
-- enterprise catalog tables
CREATE TABLE public.pg_dist_authinfo AS SELECT * FROM pg_catalog.pg_dist_authinfo;
CREATE TABLE public.pg_dist_poolinfo AS SELECT * FROM pg_catalog.pg_dist_poolinfo;
CREATE TABLE public.pg_dist_rebalance_strategy AS SELECT
name,
default_strategy,
shard_cost_function::regprocedure::text,
node_capacity_function::regprocedure::text,
shard_allowed_on_node_function::regprocedure::text,
default_threshold,
minimum_threshold,
improvement_threshold
FROM pg_catalog.pg_dist_rebalance_strategy;
-- store upgrade stable identifiers on pg_dist_object catalog
UPDATE citus.pg_dist_object
SET (type, object_names, object_args) = (SELECT * FROM pg_identify_object_as_address(classid, objid, objsubid));
END;
$cppu$;
COMMENT ON FUNCTION pg_catalog.citus_prepare_pg_upgrade()
IS 'perform tasks to copy citus settings to a location that could later be restored after pg_upgrade is done';

3
src/backend/distributed/sql/udfs/citus_prepare_pg_upgrade/latest.sql
View File

@ -40,7 +40,8 @@ BEGIN
node_capacity_function::regprocedure::text,
shard_allowed_on_node_function::regprocedure::text,
default_threshold,
minimum_threshold
minimum_threshold,
improvement_threshold
FROM pg_catalog.pg_dist_rebalance_strategy;
-- store upgrade stable identifiers on pg_dist_object catalog

35
src/backend/distributed/sql/udfs/citus_shards/10.1-1.sql generated Normal file
View File

@ -0,0 +1,35 @@
CREATE OR REPLACE VIEW pg_catalog.citus_shards AS
SELECT
pg_dist_shard.logicalrelid AS table_name,
pg_dist_shard.shardid,
shard_name(pg_dist_shard.logicalrelid, pg_dist_shard.shardid) as shard_name,
CASE WHEN partkey IS NOT NULL THEN 'distributed' WHEN repmodel = 't' THEN 'reference' ELSE 'local' END AS citus_table_type,
colocationid AS colocation_id,
pg_dist_node.nodename,
pg_dist_node.nodeport,
size as shard_size
FROM
pg_dist_shard
JOIN
pg_dist_placement
ON
pg_dist_shard.shardid = pg_dist_placement.shardid
JOIN
pg_dist_node
ON
pg_dist_placement.groupid = pg_dist_node.groupid
JOIN
pg_dist_partition
ON
pg_dist_partition.logicalrelid = pg_dist_shard.logicalrelid
LEFT JOIN
(SELECT (regexp_matches(table_name,'_(\d+)$'))[1]::int as shard_id, max(size) as size from citus_shard_sizes() GROUP BY shard_id) as shard_sizes
ON
pg_dist_shard.shardid = shard_sizes.shard_id
WHERE
pg_dist_placement.shardstate = 1
ORDER BY
pg_dist_shard.logicalrelid::text, shardid
;
GRANT SELECT ON pg_catalog.citus_shards TO public;

16
src/backend/distributed/sql/udfs/citus_shards/latest.sql
View File

@ -1,5 +1,4 @@
CREATE OR REPLACE VIEW citus.citus_shards AS
WITH shard_sizes AS (SELECT * FROM pg_catalog.citus_shard_sizes())
CREATE OR REPLACE VIEW pg_catalog.citus_shards AS
SELECT
pg_dist_shard.logicalrelid AS table_name,
pg_dist_shard.shardid,
@ -8,11 +7,7 @@ SELECT
colocationid AS colocation_id,
pg_dist_node.nodename,
pg_dist_node.nodeport,
(SELECT size FROM shard_sizes WHERE
shard_name(pg_dist_shard.logicalrelid, pg_dist_shard.shardid) = table_name
OR
'public.' || shard_name(pg_dist_shard.logicalrelid, pg_dist_shard.shardid) = table_name
LIMIT 1) as shard_size
size as shard_size
FROM
pg_dist_shard
JOIN
@ -27,9 +22,14 @@ JOIN
pg_dist_partition
ON
pg_dist_partition.logicalrelid = pg_dist_shard.logicalrelid
LEFT JOIN
(SELECT (regexp_matches(table_name,'_(\d+)$'))[1]::int as shard_id, max(size) as size from citus_shard_sizes() GROUP BY shard_id) as shard_sizes
ON
pg_dist_shard.shardid = shard_sizes.shard_id
WHERE
pg_dist_placement.shardstate = 1
ORDER BY
pg_dist_shard.logicalrelid::text, shardid
;
ALTER VIEW citus.citus_shards SET SCHEMA pg_catalog;
GRANT SELECT ON pg_catalog.citus_shards TO public;

18
src/backend/distributed/sql/udfs/get_rebalance_progress/10.1-1.sql generated Normal file
View File

@ -0,0 +1,18 @@
DROP FUNCTION pg_catalog.get_rebalance_progress();
CREATE OR REPLACE FUNCTION pg_catalog.get_rebalance_progress()
RETURNS TABLE(sessionid integer,
table_name regclass,
shardid bigint,
shard_size bigint,
sourcename text,
sourceport int,
targetname text,
targetport int,
progress bigint,
source_shard_size bigint,
target_shard_size bigint)
AS 'MODULE_PATHNAME'
LANGUAGE C STRICT;
COMMENT ON FUNCTION pg_catalog.get_rebalance_progress()
IS 'provides progress information about the ongoing rebalance operations';

17
src/backend/distributed/sql/udfs/get_rebalance_progress/9.0-1.sql generated Normal file
View File

@ -0,0 +1,17 @@
-- get_rebalance_progress returns the list of shard placement move operations along with
-- their progressions for ongoing rebalance operations.
--
CREATE OR REPLACE FUNCTION pg_catalog.get_rebalance_progress()
RETURNS TABLE(sessionid integer,
table_name regclass,
shardid bigint,
shard_size bigint,
sourcename text,
sourceport int,
targetname text,
targetport int,
progress bigint)
AS 'MODULE_PATHNAME'
LANGUAGE C STRICT;
COMMENT ON FUNCTION pg_catalog.get_rebalance_progress()
IS 'provides progress information about the ongoing rebalance operations';

18
src/backend/distributed/sql/udfs/get_rebalance_progress/latest.sql Normal file
View File

@ -0,0 +1,18 @@
DROP FUNCTION pg_catalog.get_rebalance_progress();
CREATE OR REPLACE FUNCTION pg_catalog.get_rebalance_progress()
RETURNS TABLE(sessionid integer,
table_name regclass,
shardid bigint,
shard_size bigint,
sourcename text,
sourceport int,
targetname text,
targetport int,
progress bigint,
source_shard_size bigint,
target_shard_size bigint)
AS 'MODULE_PATHNAME'
LANGUAGE C STRICT;
COMMENT ON FUNCTION pg_catalog.get_rebalance_progress()
IS 'provides progress information about the ongoing rebalance operations';

27
src/backend/distributed/sql/udfs/get_rebalance_table_shards_plan/10.1-1.sql generated Normal file
View File

@ -0,0 +1,27 @@
-- get_rebalance_table_shards_plan shows the actual events that will be performed
-- if a rebalance operation will be performed with the same arguments, which allows users
-- to understand the impact of the change overall availability of the application and
-- network trafic.
--
DROP FUNCTION pg_catalog.get_rebalance_table_shards_plan;
CREATE OR REPLACE FUNCTION pg_catalog.get_rebalance_table_shards_plan(
relation regclass default NULL,
threshold float4 default NULL,
max_shard_moves int default 1000000,
excluded_shard_list bigint[] default '{}',
drain_only boolean default false,
rebalance_strategy name default NULL,
improvement_threshold float4 DEFAULT NULL
)
RETURNS TABLE (table_name regclass,
shardid bigint,
shard_size bigint,
sourcename text,
sourceport int,
targetname text,
targetport int)
AS 'MODULE_PATHNAME'
LANGUAGE C VOLATILE;
COMMENT ON FUNCTION pg_catalog.get_rebalance_table_shards_plan(regclass, float4, int, bigint[], boolean, name, float4)
IS 'returns the list of shard placement moves to be done on a rebalance operation';

5
src/backend/distributed/sql/udfs/get_rebalance_table_shards_plan/latest.sql
View File

@ -10,7 +10,8 @@ CREATE OR REPLACE FUNCTION pg_catalog.get_rebalance_table_shards_plan(
max_shard_moves int default 1000000,
excluded_shard_list bigint[] default '{}',
drain_only boolean default false,
rebalance_strategy name default NULL
rebalance_strategy name default NULL,
improvement_threshold float4 DEFAULT NULL
)
RETURNS TABLE (table_name regclass,
shardid bigint,
@ -21,6 +22,6 @@ CREATE OR REPLACE FUNCTION pg_catalog.get_rebalance_table_shards_plan(
targetport int)
AS 'MODULE_PATHNAME'
LANGUAGE C VOLATILE;
COMMENT ON FUNCTION pg_catalog.get_rebalance_table_shards_plan(regclass, float4, int, bigint[], boolean, name)
COMMENT ON FUNCTION pg_catalog.get_rebalance_table_shards_plan(regclass, float4, int, bigint[], boolean, name, float4)
IS 'returns the list of shard placement moves to be done on a rebalance operation';

4
src/backend/distributed/test/distributed_deadlock_detection.c
View File

@ -39,6 +39,8 @@ PG_FUNCTION_INFO_V1(get_adjacency_list_wait_graph);
Datum
get_adjacency_list_wait_graph(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
TupleDesc tupleDescriptor = NULL;
HASH_SEQ_STATUS status;
@ -47,8 +49,6 @@ get_adjacency_list_wait_graph(PG_FUNCTION_ARGS)
Datum values[2];
bool isNulls[2];
CheckCitusVersion(ERROR);
Tuplestorestate *tupleStore = SetupTuplestore(fcinfo, &tupleDescriptor);
WaitGraph *waitGraph = BuildGlobalWaitGraph();
HTAB *adjacencyList = BuildAdjacencyListsForWaitGraph(waitGraph);

8
src/backend/distributed/test/foreign_key_relationship_query.c
View File

@ -77,11 +77,11 @@ drop_constraint_cascade_via_perform_deletion(PG_FUNCTION_ARGS)
Datum
get_referencing_relation_id_list(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
FuncCallContext *functionContext = NULL;
ListCell *foreignRelationCell = NULL;
CheckCitusVersion(ERROR);
/* for the first we call this UDF, we need to populate the result to return set */
if (SRF_IS_FIRSTCALL())
{
@ -136,11 +136,11 @@ get_referencing_relation_id_list(PG_FUNCTION_ARGS)
Datum
get_referenced_relation_id_list(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
FuncCallContext *functionContext = NULL;
ListCell *foreignRelationCell = NULL;
CheckCitusVersion(ERROR);
/* for the first we call this UDF, we need to populate the result to return set */
if (SRF_IS_FIRSTCALL())
{

4
src/backend/distributed/test/foreign_key_to_reference_table_rebalance.c
View File

@ -30,11 +30,11 @@ PG_FUNCTION_INFO_V1(get_foreign_key_to_reference_table_commands);
Datum
get_foreign_key_to_reference_table_commands(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
FuncCallContext *functionContext = NULL;
ListCell *commandsCell = NULL;
CheckCitusVersion(ERROR);
/* for the first we call this UDF, we need to populate the result to return set */
if (SRF_IS_FIRSTCALL())
{

4
src/backend/distributed/test/intermediate_results.c
View File

@ -34,6 +34,8 @@ PG_FUNCTION_INFO_V1(store_intermediate_result_on_node);
Datum
store_intermediate_result_on_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeNameText = PG_GETARG_TEXT_P(0);
char *nodeNameString = text_to_cstring(nodeNameText);
int nodePort = PG_GETARG_INT32(1);
@ -44,8 +46,6 @@ store_intermediate_result_on_node(PG_FUNCTION_ARGS)
bool writeLocalFile = false;
ParamListInfo paramListInfo = NULL;
CheckCitusVersion(ERROR);
WorkerNode *workerNode = FindWorkerNodeOrError(nodeNameString, nodePort);
/*

3
src/backend/distributed/test/metadata_sync.c
View File

@ -15,6 +15,7 @@
#include "catalog/pg_type.h"
#include "distributed/connection_management.h"
#include "distributed/intermediate_result_pruning.h"
#include "distributed/listutils.h"
#include "distributed/maintenanced.h"
#include "distributed/metadata_sync.h"
@ -104,7 +105,7 @@ wait_until_metadata_sync(PG_FUNCTION_ARGS)
}
MultiConnection *connection = GetNodeConnection(FORCE_NEW_CONNECTION,
"localhost", PostPortNumber);
LOCAL_HOST_NAME, PostPortNumber);
ExecuteCriticalRemoteCommand(connection, "LISTEN " METADATA_SYNC_CHANNEL);
int waitFlags = WL_SOCKET_READABLE | WL_TIMEOUT | WL_POSTMASTER_DEATH;

6
src/backend/distributed/test/run_from_same_connection.c
View File

@ -17,10 +17,10 @@
#include "access/xact.h"
#include "distributed/connection_management.h"
#include "distributed/coordinator_protocol.h"
#include "distributed/function_utils.h"
#include "distributed/intermediate_result_pruning.h"
#include "distributed/lock_graph.h"
#include "distributed/coordinator_protocol.h"
#include "distributed/metadata_cache.h"
#include "distributed/remote_commands.h"
#include "distributed/run_from_same_connection.h"
@ -83,13 +83,13 @@ AllowNonIdleTransactionOnXactHandling(void)
Datum
start_session_level_connection_to_node(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
text *nodeName = PG_GETARG_TEXT_P(0);
uint32 nodePort = PG_GETARG_UINT32(1);
char *nodeNameString = text_to_cstring(nodeName);
int connectionFlags = 0;
CheckCitusVersion(ERROR);
if (singleConnection != NULL && (strcmp(singleConnection->hostname,
nodeNameString) != 0 ||
singleConnection->port != nodePort))

15
src/backend/distributed/test/shard_rebalancer.c
View File

@ -21,6 +21,7 @@
#include "distributed/connection_management.h"
#include "distributed/listutils.h"
#include "distributed/multi_physical_planner.h"
#include "distributed/shard_cleaner.h"
#include "distributed/shard_rebalancer.h"
#include "funcapi.h"
#include "miscadmin.h"
@ -50,6 +51,7 @@ static ShardCost GetShardCost(uint64 shardId, void *context);
PG_FUNCTION_INFO_V1(shard_placement_rebalance_array);
PG_FUNCTION_INFO_V1(shard_placement_replication_array);
PG_FUNCTION_INFO_V1(worker_node_responsive);
PG_FUNCTION_INFO_V1(run_try_drop_marked_shards);
typedef struct ShardPlacementTestInfo
{
@ -71,6 +73,17 @@ typedef struct RebalancePlanContext
List *shardPlacementTestInfoList;
} RebalancePlacementContext;
/*
* run_try_drop_marked_shards is a wrapper to run TryDropOrphanedShards.
*/
Datum
run_try_drop_marked_shards(PG_FUNCTION_ARGS)
{
bool waitForLocks = false;
TryDropOrphanedShards(waitForLocks);
PG_RETURN_VOID();
}
/*
* shard_placement_rebalance_array returns a list of operations which can make a
@ -89,6 +102,7 @@ shard_placement_rebalance_array(PG_FUNCTION_ARGS)
float threshold = PG_GETARG_FLOAT4(2);
int32 maxShardMoves = PG_GETARG_INT32(3);
bool drainOnly = PG_GETARG_BOOL(4);
float utilizationImproventThreshold = PG_GETARG_FLOAT4(5);
List *workerNodeList = NIL;
List *shardPlacementListList = NIL;
@ -143,6 +157,7 @@ shard_placement_rebalance_array(PG_FUNCTION_ARGS)
threshold,
maxShardMoves,
drainOnly,
utilizationImproventThreshold,
&rebalancePlanFunctions);
ArrayType *placementUpdateJsonArray = PlacementUpdateListToJsonArray(
placementUpdateList);

15
src/backend/distributed/transaction/backend_data.c
View File

@ -106,6 +106,8 @@ PG_FUNCTION_INFO_V1(get_all_active_transactions);
Datum
assign_distributed_transaction_id(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid userId = GetUserId();
/* prepare data before acquiring spinlock to protect against errors */
@ -113,8 +115,6 @@ assign_distributed_transaction_id(PG_FUNCTION_ARGS)
uint64 transactionNumber = PG_GETARG_INT64(1);
TimestampTz timestamp = PG_GETARG_TIMESTAMPTZ(2);
CheckCitusVersion(ERROR);
/* MyBackendData should always be avaliable, just out of paranoia */
if (!MyBackendData)
{
@ -166,14 +166,14 @@ assign_distributed_transaction_id(PG_FUNCTION_ARGS)
Datum
get_current_transaction_id(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
TupleDesc tupleDescriptor = NULL;
Datum values[5];
bool isNulls[5];
CheckCitusVersion(ERROR);
/* build a tuple descriptor for our result type */
if (get_call_result_type(fcinfo, NULL, &tupleDescriptor) != TYPEFUNC_COMPOSITE)
{
@ -225,12 +225,13 @@ get_current_transaction_id(PG_FUNCTION_ARGS)
Datum
get_global_active_transactions(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
TupleDesc tupleDescriptor = NULL;
List *workerNodeList = ActivePrimaryNonCoordinatorNodeList(NoLock);
List *connectionList = NIL;
StringInfo queryToSend = makeStringInfo();
CheckCitusVersion(ERROR);
Tuplestorestate *tupleStore = SetupTuplestore(fcinfo, &tupleDescriptor);
appendStringInfo(queryToSend, GET_ACTIVE_TRANSACTION_QUERY);
@ -336,9 +337,9 @@ get_global_active_transactions(PG_FUNCTION_ARGS)
Datum
get_all_active_transactions(PG_FUNCTION_ARGS)
{
TupleDesc tupleDescriptor = NULL;
CheckCitusVersion(ERROR);
TupleDesc tupleDescriptor = NULL;
Tuplestorestate *tupleStore = SetupTuplestore(fcinfo, &tupleDescriptor);
StoreAllActiveTransactions(tupleStore, tupleDescriptor);

3
src/backend/distributed/transaction/relation_access_tracking.c
View File

@ -793,7 +793,8 @@ CheckConflictingRelationAccesses(Oid relationId, ShardPlacementAccessType access
"foreign keys. Any parallel modification to "
"those hash distributed tables in the same "
"transaction can only be executed in sequential query "
"execution mode", relationName)));
"execution mode",
relationName != NULL ? relationName : "<dropped>")));
/*
* Switching to sequential mode is admittedly confusing and, could be useless

6
src/backend/distributed/transaction/transaction_management.c
View File

@ -101,7 +101,7 @@ MemoryContext CommitContext = NULL;
* do 2PC on the remote connections that did a modification.
*
* As a variable name ShouldCoordinatedTransactionUse2PC could
* be improved. We use CoordinatedTransactionShouldUse2PC() as the
* be improved. We use Use2PCForCoordinatedTransaction() as the
* public API function, hence couldn't come up with a better name
* for the underlying variable at the moment.
*/
@ -190,14 +190,14 @@ InCoordinatedTransaction(void)
/*
* CoordinatedTransactionShouldUse2PC() signals that the current coordinated
* Use2PCForCoordinatedTransaction() signals that the current coordinated
* transaction should use 2PC to commit.
*
* Note that even if 2PC is enabled, it is only used for connections that make
* modification (DML or DDL).
*/
void
CoordinatedTransactionShouldUse2PC(void)
Use2PCForCoordinatedTransaction(void)
{
Assert(InCoordinatedTransaction());

4
src/backend/distributed/transaction/worker_transaction.c
View File

@ -96,7 +96,7 @@ SendCommandToWorkerAsUser(const char *nodeName, int32 nodePort, const char *node
uint32 connectionFlags = 0;
UseCoordinatedTransaction();
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
MultiConnection *transactionConnection = GetNodeUserDatabaseConnection(
connectionFlags, nodeName,
@ -404,7 +404,7 @@ SendCommandToWorkersParamsInternal(TargetWorkerSet targetWorkerSet, const char *
List *workerNodeList = TargetWorkerSetNodeList(targetWorkerSet, ShareLock);
UseCoordinatedTransaction();
CoordinatedTransactionShouldUse2PC();
Use2PCForCoordinatedTransaction();
/* open connections in parallel */
WorkerNode *workerNode = NULL;

1
src/backend/distributed/utils/citus_copyfuncs.c
View File

@ -135,6 +135,7 @@ CopyNodeDistributedPlan(COPYFUNC_ARGS)
COPY_NODE_FIELD(subPlanList);
COPY_NODE_FIELD(usedSubPlanNodeList);
COPY_SCALAR_FIELD(fastPathRouterPlan);
COPY_SCALAR_FIELD(numberOfTimesExecuted);
COPY_NODE_FIELD(planningError);
}

1
src/backend/distributed/utils/citus_outfuncs.c
View File

@ -198,6 +198,7 @@ OutDistributedPlan(OUTFUNC_ARGS)
WRITE_NODE_FIELD(subPlanList);
WRITE_NODE_FIELD(usedSubPlanNodeList);
WRITE_BOOL_FIELD(fastPathRouterPlan);
WRITE_UINT_FIELD(numberOfTimesExecuted);
WRITE_NODE_FIELD(planningError);
}

10
src/backend/distributed/utils/colocation_utils.c
View File

@ -70,6 +70,9 @@ PG_FUNCTION_INFO_V1(update_distributed_table_colocation);
Datum
mark_tables_colocated(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
Oid sourceRelationId = PG_GETARG_OID(0);
ArrayType *relationIdArrayObject = PG_GETARG_ARRAYTYPE_P(1);
@ -80,8 +83,6 @@ mark_tables_colocated(PG_FUNCTION_ARGS)
"operation")));
}
CheckCitusVersion(ERROR);
EnsureCoordinator();
EnsureTableOwner(sourceRelationId);
Datum *relationIdDatumArray = DeconstructArrayObject(relationIdArrayObject);
@ -108,11 +109,12 @@ mark_tables_colocated(PG_FUNCTION_ARGS)
Datum
update_distributed_table_colocation(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
EnsureCoordinator();
Oid targetRelationId = PG_GETARG_OID(0);
text *colocateWithTableNameText = PG_GETARG_TEXT_P(1);
CheckCitusVersion(ERROR);
EnsureCoordinator();
EnsureTableOwner(targetRelationId);
char *colocateWithTableName = text_to_cstring(colocateWithTableNameText);

8
src/backend/distributed/utils/distribution_column.c
View File

@ -49,12 +49,12 @@ PG_FUNCTION_INFO_V1(column_to_column_name);
Datum
column_name_to_column(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *columnText = PG_GETARG_TEXT_P(1);
char *columnName = text_to_cstring(columnText);
CheckCitusVersion(ERROR);
Relation relation = relation_open(relationId, AccessShareLock);
Var *column = BuildDistributionKeyFromColumnName(relation, columnName);
@ -100,13 +100,13 @@ column_name_to_column_id(PG_FUNCTION_ARGS)
Datum
column_to_column_name(PG_FUNCTION_ARGS)
{
CheckCitusVersion(ERROR);
Oid relationId = PG_GETARG_OID(0);
text *columnNodeText = PG_GETARG_TEXT_P(1);
char *columnNodeString = text_to_cstring(columnNodeText);
CheckCitusVersion(ERROR);
char *columnName = ColumnToColumnName(relationId, columnNodeString);
text *columnText = cstring_to_text(columnName);

10
src/backend/distributed/utils/foreign_key_relationship.c
View File

@ -32,6 +32,7 @@
#if PG_VERSION_NUM >= PG_VERSION_13
#include "common/hashfn.h"
#endif
#include "utils/inval.h"
#include "utils/memutils.h"
@ -82,6 +83,7 @@ static ForeignConstraintRelationshipNode * GetRelationshipNodeForRelationId(Oid
relationId,
bool *isFound);
static void CreateForeignConstraintRelationshipGraph(void);
static bool IsForeignConstraintRelationshipGraphValid(void);
static List * GetNeighbourList(ForeignConstraintRelationshipNode *relationshipNode,
bool isReferencing);
static List * GetRelationIdsFromRelationshipNodeList(List *fKeyRelationshipNodeList);
@ -348,9 +350,15 @@ CreateForeignConstraintRelationshipGraph()
/*
* IsForeignConstraintGraphValid check whether there is a valid graph.
*/
bool
static bool
IsForeignConstraintRelationshipGraphValid()
{
/*
* We might have some concurrent metadata changes. In order to get the changes,
* we first need to accept the cache invalidation messages.
*/
AcceptInvalidationMessages();
if (fConstraintRelationshipGraph != NULL && fConstraintRelationshipGraph->isValid)
{
return true;

6
src/backend/distributed/utils/maintenanced.c
View File

@ -93,7 +93,7 @@ typedef struct MaintenanceDaemonDBData
/* config variable for distributed deadlock detection timeout */
double DistributedDeadlockDetectionTimeoutFactor = 2.0;
int Recover2PCInterval = 60000;
int DeferShardDeleteInterval = 60000;
int DeferShardDeleteInterval = 15000;
/* config variables for metadata sync timeout */
int MetadataSyncInterval = 60000;
@ -644,8 +644,8 @@ CitusMaintenanceDaemonMain(Datum main_arg)
*/
lastShardCleanTime = GetCurrentTimestamp();
bool waitForCleanupLock = false;
numberOfDroppedShards = TryDropMarkedShards(waitForCleanupLock);
bool waitForLocks = false;
numberOfDroppedShards = TryDropOrphanedShards(waitForLocks);
}
CommitTransactionCommand();

2
src/backend/distributed/utils/reference_table_utils.c
View File

@ -193,7 +193,7 @@ EnsureReferenceTablesExistOnAllNodesExtended(char transferMode)
int connectionFlags = OUTSIDE_TRANSACTION;
MultiConnection *connection = GetNodeUserDatabaseConnection(
connectionFlags, "localhost", PostPortNumber,
connectionFlags, LocalHostName, PostPortNumber,
userName, NULL);
if (PQstatus(connection->pgConn) == CONNECTION_OK)

Some files were not shown because too many files have changed in this diff Show More