(Hopefully) Fixes#5000.
If memory allocation done for `SubXactContext *state` in `PushSubXact()`
fails, then `PopSubXact()` might segfault, for example, when grabbing
the
topmost `SubXactContext` from `activeSubXactContexts` if this is the
first
ever subxact within the current xact, with the following stack trace:
```c
citus.so!list_nth_cell(const List * list, int n) (\opt\pgenv\pgsql-14.3\include\server\nodes\pg_list.h:260)
citus.so!PopSubXact(SubTransactionId subId) (\home\onurctirtir\citus\src\backend\distributed\transaction\transaction_management.c:761)
citus.so!CoordinatedSubTransactionCallback(SubXactEvent event, SubTransactionId subId, SubTransactionId parentSubid, void * arg) (\home\onurctirtir\citus\src\backend\distributed\transaction\transaction_management.c:673)
CallSubXactCallbacks(SubXactEvent event, SubTransactionId mySubid, SubTransactionId parentSubid) (\opt\pgenv\src\postgresql-14.3\src\backend\access\transam\xact.c:3644)
AbortSubTransaction() (\opt\pgenv\src\postgresql-14.3\src\backend\access\transam\xact.c:5058)
AbortCurrentTransaction() (\opt\pgenv\src\postgresql-14.3\src\backend\access\transam\xact.c:3366)
PostgresMain(int argc, char ** argv, const char * dbname, const char * username) (\opt\pgenv\src\postgresql-14.3\src\backend\tcop\postgres.c:4250)
BackendRun(Port * port) (\opt\pgenv\src\postgresql-14.3\src\backend\postmaster\postmaster.c:4530)
BackendStartup(Port * port) (\opt\pgenv\src\postgresql-14.3\src\backend\postmaster\postmaster.c:4252)
ServerLoop() (\opt\pgenv\src\postgresql-14.3\src\backend\postmaster\postmaster.c:1745)
PostmasterMain(int argc, char ** argv) (\opt\pgenv\src\postgresql-14.3\src\backend\postmaster\postmaster.c:1417)
main(int argc, char ** argv) (\opt\pgenv\src\postgresql-14.3\src\backend\main\main.c:209)
```
For this reason, to be more defensive against memory-allocation errors
that could happen at `PushSubXact()`, now we use our pre-allocated
memory
context for the objects created in `PushSubXact()`.
This commit also attempts reducing the memory allocations done under
CommitContext to reduce the chances of consuming all the memory
available
to CommitContext.
Note that it's problematic to encounter with such a memory-allocation
error for other objects created in `PushSubXact()` as well, so above is
an **example** scenario that might result in a segfault.
DESCRIPTION: Fixes a bug that might cause segfaults when handling deeply
nested subtransactions
DESCRIPTION: Makes sure to disallow triggers that depend on extensions
We were already doing so for `ALTER trigger DEPENDS ON EXTENSION`
commands. However, we also need to disallow creating Citus tables
having such triggers already, so this PR fixes that.
DESCRIPTION: Improve a query that terminates compeling backends from citus_update_node()
1. Use pg_blocking_pids() function instead of self join on pg_locks. It exists since 9.6 and more accurate than pg_locks.
2. Prefix all function calls with pg_catalog schema to prevent privilege escalation by creating functions with similar names in a public schema.
3. Change logs and update comments to reflect the fact that the pg_terminate_backend() function only sends SIGTERM but not wating for the actual backend termination.
DESCRIPTION: Allow citus_update_node() to work with nodes from different clusters
citus_update_node(), citus_nodename_for_nodeid(), and citus_nodeport_for_nodeid() functions only checked for nodes in their own clusters and hence last two returned NULLs and the first one showed an error is the nodeId was from a different cluster.
Fixes https://github.com/citusdata/citus/issues/6433
increasing logical clock. Clock guarantees to never go back in value after restarts,
and makes best attempt to keep the value close to unix epoch time in milliseconds.
Also, introduces a new GUC "citus.enable_cluster_clock", when true, every
distributed transaction is stamped with logical causal clock and persisted
in a catalog pg_dist_commit_transaction.
DESCRIPTION: Drops GUC defer_drop_after_shard_split
DESCRIPTION: Drops GUC defer_drop_after_shard_move
Drop GUCs and related parts from the code.
Delete tests that specifically added for the GUCs.
Keep tests that can be used without the GUCs.
Update test output changes.
The motivation for this PR is to have an "always deferring" mechanism.
These two GUCs provide an option to not deferring dropping objects
during a shard move/split, and dropping them immediately. With this PR,
we will be always deferring dropping orphaned shards and other types of
objects.
We will have a separate PR to extend the deferred cleanup operation, so
that we would create records for deferred drop, for Subscriptions,
Publications, Replication Slots etc. This will make us be able to keep
track of created objects that needs to be dropped, during a shard
move/split. We will have objects created specifically for the current
operation; and those objects will be dropped at the end.
We have an issue (a draft roadmap) for enabling parallel shard moves.
For details please see: https://github.com/citusdata/citus/issues/6437
This didn't cause any bugs since today we're always calling
UpdateAutoConvertedForConnectedRelations with autoconverted=false, so we
don't need to backport this to anywhere.
We should not introduce breaking sql changes to upgrade files after they
are released. We did that for worker_fetch_foreign_file in v9.0.0 and
worker_repartition_cleanup in v9.2.0. Later when we try to drop those
udfs, they were missing for some clients unexpectedly due to breaking
change in an old upgrade script. For that case, the fix is to add DROP
IF EXISTS for those 2 udfs in 11.0-4--11.1-1.
This crash happens with recursively planned queries. For such queries,
subplans are explained via the ExplainOnePlan function of postgresql.
This function reconstructs the query description from the plan therefore
it expects the ActiveSnaphot for the query be available. This fix makes
sure that the snapshot is in the stack before calling ExplainOnePlan.
Fixes#2920.
DESCRIPTION: Don't leak search_path to workers on DDL
For DDL we have to set the `search_path` on workers to the same as on
the coordinator for some DDL to work. Previously this search_path would
leak outside of the transaction that was used for the DDL. This fixes
that by using `SET LOCAL` instead of `SET`. The only place where we
still use plain `SET` is for DDL commands that are not allowed within
transactions, such as `CREATE INDEX CONCURRENLTY`.
This fixes this flaky test:
```diff
CONTEXT: SQL statement "SELECT change_id FROM distributed_triggers.data_changes
WHERE shard_key_value = NEW.shard_key_value AND object_id = NEW.object_id
ORDER BY change_id DESC LIMIT 1"
-PL/pgSQL function record_change() line XX at SQL statement
+PL/pgSQL function distributed_triggers.record_change() line 17 at SQL statement
while executing command on localhost:57638
DELETE FROM data_ref_table where shard_key_value = 'hello';
```
Source:
https://app.circleci.com/pipelines/github/citusdata/citus/27849/workflows/75ae5f1a-100b-4b7a-b991-7de069f39ee1/jobs/831429
I had tried to fix this flaky test in #5894 and then I tried
implementing a better fix in #5896, where @marcocitus suggested this
better fix. This change reverts the fix from #5894 and implements the
fix suggested by Marco.
Our multi_mx_alter_distributed_table test actually depended on the old
buggy search_path leaking behavior. After fixing the bug that test would
fail like this:
```diff
CALL proc_0(1.0);
DEBUG: pushing down the procedure
-NOTICE: Res: 3
-DETAIL: from localhost:xxxxx
+ERROR: relation "test_proc_colocation_0" does not exist
+CONTEXT: PL/pgSQL function mx_alter_distributed_table.proc_0(double precision) line 5 at SQL statement
+while executing command on localhost:57637
RESET client_min_messages;
```
I fixed this test by fully qualifying the table names used in the
procedure. I think it's quite unlikely that actual users depend
on this behavior though. Since it would require first doing
DDL before calling a procedure in a session where the
search_path was changed after connecting.
DESCRIPTION: Adds failure test for shard move
DESCRIPTION: Remove function `WaitForAllSubscriptionsToBecomeReady` and
related tests
Adding some failure tests for shard moves.
Dropping the not-needed-anymore function
`WaitForAllSubscriptionsToBecomeReady`, as the subscriptions now start
as ready from the beginning because we don't use logical replication
table sync workers anymore.
fixes: #6260
To be able to test non-blocking shard moves we take an advisory lock, so
we can pause the shard move at an interesting moment. Originally this
was during the logical replication catch up phase. But when I added
tests for the rebalancer progress I moved this lock before the initial
data copy. This allowed testing of the rebalance progress, but
inadvertently made our non-blocking tests not actually test if we held
unintended locks during logical replication catch up.
This fixes that by creating two types of advisory locks, one before the
copy and one after. This causes the tests to actually test their
intended scenario again.
Furthermore it starts using one of these locks for blocking shard moves
too. Which allowed me to reduce the complexity of the rebalance progress
test suite quite a bit. It also allowed enabling some flaky tests again,
because this stopped them from being flaky. And finally it allowed
testing of rebalance progress for blocking shard copy operations as
well.
In passing it fixes a flaky test during parallel blocking shard moves by
ordering the output.
DESCRIPTION: Adds status column to get_rebalance_progress()
Introduces a new column named `status` for the function
`get_rebalance_progress()`. For each ongoing shard move, this column
will reveal information about that shard move operation's current
status.
For now, candidate status messages could be one of the below.
* Not Started
* Setting Up
* Copying Data
* Catching Up
* Creating Constraints
* Final Catchup
* Creating Foreign Keys
* Completing
* Completed
Deparser function set_relation_column_names() knows that it needs to
re-evaluate column names based on relation's tuple descriptor when
the rte belongs to a relation (RTE_RELATION).
However before this commit, it didn't know about the fact that citus
might wrap such an rte with an rte that points to
citus_extradata_container() placeholder.
And because of this, it was simply taking the column aliases
(e.g., "bar" in "foo AS bar") into the account and this might result in
an incorrectly deparsed query as in below case:
* Say, if we had view based on following query:
```sql
SELECT a FROM table;
```
* And if we rename column "a" to "b", the view query normally becomes:
```sql
SELECT b AS a FROM table;
```
* So before this commit, deparsing a query based on that view was
resulting in such a query due to deparsing based on the column aliases,
which is not correct:
```sql
SELECT a FROM table;
```
Fixes#5932.
DESCRIPTION: Fixes a bug that might cause failing to query the views
based on tables that have renamed columns
PostgreSQL 15 exposes WL_SOCKET_CLOSED in WaitEventSet API, which is
useful for detecting closed remote sockets. In this patch, we use this
new event and try to detect closed remote sockets in the executor.
When a closed socket is detected, the executor now has the ability to
retry the connection establishment. Note that, the executor can retry
connection establishments only for the connection that has not been
used. Basically, this patch is mostly useful for preventing the executor
to fail if a cached connection is closed because of the worker node
restart (or worker failover).
In other words, the executor cannot retry connection establishment if we
are in a distributed transaction AND any command has been sent over the
connection. That requires more sophisticated retry mechanisms. For now,
fixing the above use case is enough.
Fixes#5538
Earlier discussions: #5908, #6259 and #6283
### Summary of the current approach regards to earlier trials
As noted, we explored some alternatives before getting into this.
https://github.com/citusdata/citus/pull/6283 is simple, but lacks an
important property. We should be checking for `WL_SOCKET_CLOSED`
_before_ sending anything over the wire. Otherwise, it becomes very
tricky to understand which connection is actually safe to retry. For
example, in the current patch, we can safely check
`transaction->transactionState == REMOTE_TRANS_NOT_STARTED` before
restarting a connection.
#6259 does what we intent here (e.g., check for sending any command).
However, as @marcocitus noted, it is very tricky to handle
`WaitEventSets` in multiple places. And, the executor is designed such
that it reacts to the events. So, adding anything `pre-executor` seemed
too ugly.
In the end, I converged into this patch. This patch relies on the
simplicity of #6283 and also does a very limited handling of
`WaitEventSets`, just for our purpose. Just before we add any connection
to the execution, we check if the remote session has already closed.
With that, we do a brief interaction of multiple wait event processing,
but with different purposes. The new wait event processing we added does
not even consider cancellations. We let that handled by the main event
processing loop.
Co-authored-by: Marco Slot <marco.slot@gmail.com>
In #6405 I added better improved blocked process detection for isolation
tests. But when cleaning up unnecessary code I cleaned up a bit too
much. This actually includes the new function definition in our
migrations.
If an operation requires having coordinator in pg_dist_node and if that
is not the case, then we automatically add the coordinator into
pg_dist_node if user didn't add any worker nodes yet.
However, if user have already added some worker nodes before, we throw
an error. With this commit, we improve the error thrown in that case.
Closes#6423 based on the discussion made there.
Sometimes our CI randomly fails on a test in a way similar to this:
```diff
step s2-drop:
DROP TABLE cancel_table;
-
+ <waiting ...>
+step s2-drop: <... completed>
starting permutation: s1-timeout s1-begin s1-sleep10000 s1-rollback s1-reset s1-drop
```
Source:
https://app.circleci.com/pipelines/github/citusdata/citus/26524/workflows/5415b84f-13a3-482f-bef9-648314c79a67/jobs/756377
I tried to fix that already in #6252 by disabling the maintenance daemon
during isolation tests. But it seems that hasn't fixed all cases of
these errors. This is another attempt at fixing these issues that seems
to have better results.
What it does is that it starts using the pInterestingPids parameter that
citus_isolation_test_session_is_blocked receives. With this change we
start filter out block-edges that are not caused by any of these pids.
In passing this change also makes it possible to run
`isolation_create_distributed_table_concurrently` with
`check-isolation-base`
PG15 introduced a function called ReplicationSlotName that causes
conflicts with our function with the same name. I solved this issue by
renaming our function to ReplicationSlotNameForNodeAndOwner
Relevant PG commit:
c3b5992b91
DESCRIPTION: Fix bug in global PID assignment for rebalancer
sub-connections
In CI our isolation_shard_rebalancer_progress test would sometimes fail
like this:
```diff
+isolationtester: canceling step s1-rebalance-c1-block-writes after 60 seconds
step s1-rebalance-c1-block-writes:
SELECT rebalance_table_shards('colocated1', shard_transfer_mode:='block_writes');
- <waiting ...>
+
+ERROR: canceling statement due to user request
step s7-get-progress:
```
Source:
https://app.circleci.com/pipelines/github/citusdata/citus/27855/workflows/2a7e335a-f3e8-46ed-b6bd-6920d42f7214/jobs/831710
It turned out this was an actual bug in the way our assigning of global
PIDs interacts with the way we connect to ourselves as the shard
rebalancer. The first command the shard rebalancer sends is a SET
ommand to change the application_name to `citus_rebalancer`. If
`StartupCitusBackend` is called after this command is processed, then it
overwrites the global PID that was extracted from the previous
application_name. This makes sure that we don't do that, and continue to
use the original global PID. While it might seem that we only call
`StartupCitusBackend` once for each query backend, this isn't actually
the case. Whenever pg_dist_partition gets ANALYZEd by autovacuum
we indirectly call `StartupCitusBackend` again, because we invalidate
the cache then.
In passing this fixes two other things as well:
1. It sets `distributedCommandOriginator` correctly in
`AssignGlobalPID`, by using IsExternalClientBackend(). This doesn't
matter much anymore, since AssignGlobalPID effectively becomes a
no-op in this PR for any non-external client backends.
2. It passes the application_name to InitializeBackendData in
StartupCitusBackend, instead of INVALID_CITUS_INTERNAL_BACKEND_GPID
(which effectively got casted to NULL). In practice this doesn't
change the behaviour of the call, since the call is a no-op for every
backend except the maintenance daemon. And the behaviour of the call
is the same for NULL as for the application_name of the maintenance
daemon.
Fixes https://github.com/citusdata/citus/issues/6394.
DESCRIPTION: Fixes a bug that causes creating disabled-triggers on
shards as enabled
Since CREATE TRIGGER doesn't have syntax support to specify
whether the trigger should be enabled/disabled, the underlying
PG function (`pg_get_triggerdef()`) that we use to generate the
command to create the trigger is not enough. For this reason, we
append a second command to enable/disable trigger, right after
creating it.
We don't retain explicit extension dependencies set by using
`ALTER trigger DEPENDS ON EXTENSION` commands too, but apparently
right fix for that is to throw an error as in
`PreprocessAlterTriggerDependsStmt()`; so, opened a separate PR
to fix that #6399.
During alter_distributed_table, we create a new table like the
original table but with the altered options.
To retrieve the name of the distribution column, we were using
the attribute syscache of the new table, since we already created
the new table as identical to the original table.
However, the attribute syscaches of these two tables are not
the same if the original table has dropped columns. The reason
is that dropped columns are all still present in the cache.
Hence, for example, the attnos would be different in the syscaches.
So, let's use the attribute syscache of the original table.
DESCRIPTION: Adds source_lsn and target_lsn fields into
get_rebalance_progress
Adding two fields named `source_lsn` and `target_lsn` to the function
`get_rebalance_progress`.
Target lsn data is fetched in `GetShardStatistics`, by expanding the
query sent to workers (joining with pg_subscription_rel and
pg_stat_subscription). Then put into the hashmap, for each shard.
Source lsn data is fetched in `BuildWorkerShardStatististicsHash`, in
the loop that iterate each node, by sending a pg_current_wal_lsn query
to each node. Then put into the hashmap, for each node.
In Split, Logical replication logic and ShardCleaner we call
`SendCommandListToWorkerOutsideTransaction` and
`SendOptionalCommandListToWorkerOutsideTransaction` frequently. This
opens new connection for each of those calls, even though we already
have a perfectly good connection lying around.
This PR adds two new APIs
`SendCommandListToWorkerOutsideTransactionWithConnection` and
`SendOptionalCommandListToWorkerOutsideTransactionWithConnection` that
allow sending a list of queries in a transaction over an existing
connection. We also update the callers (Split, ShardCleaner, Logical
Replication) to use these new APIs instead.
Co-authored-by: Nitish Upreti <niupre@microsoft.com>
Co-authored-by: Onder Kalaci <onderkalaci@gmail.com>
In Citus 11.1.0 we changed the order of doing the initial data copy and
the replica identity creation when doing a non blocking shard move. This
was done to try and increase the speed with which shard moves could be
done. But after doing more extensive performance testing this change
turned out to have a negative impact on the speed of moves on the setups
that I tested.
Looking at the resource usage metrics of the VMs the reason for this
seems to be that these shard moves were bottlenecked by disk bandwidth.
While creating replica identities in bulk after the initial copy will
reduce CPU usage a bit, it does require an additional sequence scan of
the just written data. So when a VM is bottlenecked on disk, it makes
sense to spend a little bit more CPU to avoid an additional scan. Since
PKs are usually simple indexes that don't require lots of CPU to update,
as opposed to e.g. GiST indexes.
This reverts the order change to avoid a regression on shard move speed
in these cases.
For future releases we might consider re-evaluating our index creation
order for other indexes too, and create "simple" indexes before the
copy.
Given that we drop DEFAULT nextval('sequence') expressions from
shard relation columns, allowing `ON DELETE/UPDATE SET DEFAULT`
on such columns might cause inserting NULL values as a result
of a delete/update operation.
For this reason, we disallow ON DELETE/UPDATE SET DEFAULT actions
on columns that default to sequences.
DESCRIPTION: Disallows having ON DELETE/UPDATE SET DEFAULT actions on
columns that default to sequences
Fixes#6339.
As we did for GENERATED STORED columns in #4613, we should not drop
column
default expressions that are not based on sequences from shard relation
since
such expressions need to exist e.g. for foreign key actions.
For the column default expressions that are based on sequences we cannot
do much, so we need to disallow having ON DELETE SET DEFAULT actions on
such columns in a separate PR, see #6339.
Fixes#6318.
DESCRIPTION: Fixes a bug that might cause inserting incorrect DEFAULT
values when applying foreign key actions
DESCRIPTION: Fixes dropping replication slots
As detected by a flaky test, Citus sometimes fails to drop replication
slots, possibly due to a race condition, at the end of a shard split.
With this PR, we retry to drop them in case of an `OBJECT_IN_USE` error,
consistently for 20 seconds.
fixes: #6326
DESCRIPTION: Improve logging during shard split and resource cleanup
### DESCRIPTION
This PR makes logging improvements to Shard Split :
1. Update confusing logging to fix#6312
2. Added new `ereport(LOG` to make debugging easier as part of telemetry review.
Comment from the code is clear on this:
/*
* The statistics objects of the distributed table are not relevant
* for the distributed planning, so we can override it.
*
* Normally, we should not need this. However, the combination of
* Postgres commit 269b532aef55a579ae02a3e8e8df14101570dfd9 and
* Citus function AdjustPartitioningForDistributedPlanning()
* forces us to do this. The commit expects statistics objects
* of partitions to have "inh" flag set properly. Whereas, the
* function overrides "inh" flag. To avoid Postgres to throw error,
* we override statlist such that Postgres does not try to process
* any statistics objects during the standard_planner() on the
* coordinator. In the end, we do not need the standard_planner()
* on the coordinator to generate an optimized plan. We call
* into standard_planner() for other purposes, such as generating the
* relationRestrictionContext here.
*
* AdjustPartitioningForDistributedPlanning() is a hack that we use
* to prevent Postgres' standard_planner() to expand all the partitions
* for the distributed planning when a distributed partitioned table
* is queried. It is required for both correctness and performance
* reasons. Although we can eliminate the use of the function for
* the correctness (e.g., make sure that rest of the planner can handle
* partitions), it's performance implication is hard to avoid. Certain
* planning logic of Citus (such as router or query pushdown) relies
* heavily on the relationRestrictionList. If
* AdjustPartitioningForDistributedPlanning() is removed, all the
* partitions show up in the, causing high planning times for
* such queries.
*/
DESCRIPTION: Fixes floating exception during
create_distributed_table_concurrently.
Fixes#6332.
During create_distributed_table_concurrently, when there is no active
primary node, it fails with floating exception. We added similar check
with create_distributed_table. It will fail with proper message if
current active node is less than replication factor.
The PR introduces code changes to fix Issue
[6303](https://github.com/citusdata/citus/issues/6303)
`create_distributed_table_concurrently` following drop column, creates a
buggy situation in split decoder.
* Consider the below scenario:
* Session1 : Drop column followed by
create_distributed_table_concurrently
* Session2 : Concurrent insert workload
The child shards created by `create_distributed_table_concurrently` will
have less columns than the source shard because some column were
dropped. The incoming tuple from session2 will have more columns as the
writes happened on source shard. But now the tuple needs to be applied
on child shard. So we need to format existing tuple according to child
schema and skip dropped column values.
The PR fixes this by reformatting the tuple according the target child
schema.
Test:
1) isolation_create_distributed_concurrently_after_drop_column - Repros
the issue and tests on the same.
No need for description, fixing issue introduced with new feature for
11.1
Fixes#6333
Due to Postgres' C api being o-indexed and postgres' attributes being
1-indexed, we were reading the wrong Datum as the Task owner when
cancelling. Here we add a test to show the error and fix the off-by-one
error.
When I built Citus on PG15beta4 locally, I get a warning message.
```
utils/background_jobs.c:902:5: warning: declaration does not declare anything
[-Wmissing-declarations]
__attribute__((fallthrough));
^
1 warning generated.
```
This is a hint to the compiler that we are deliberately falling through
in a switch-case block.
DESCRIPTION: Show citus_copy_shard_placement progress in
get_rebalance_progress
When rebalancing to a new node that does not have reference tables yet
the rebalancer will first copy the reference tables to the nodes.
Depending on the size of the reference tables, this might take a long
time. However, there's no indication of what's happening at this stage
of the rebalance.
This PR improves this situation by also showing the progress of any
citus_copy_shard_placement calls when calling get_rebalance_progress.
We can now do the following:
- Distribute sequence with logged/unlogged option
- ALTER TABLE my_sequence SET LOGGED/UNLOGGED
- ALTER SEQUENCE my_sequence SET LOGGED/UNLOGGED
Relevant PG commit
344d62fb9a
PG15 introduces `CLUSTER` commands for partitioned tables. Similar to a
`CLUSTER` command with no supplied table names, these commands also can
not be run inside transaction blocks and therefore can not be propagated
in a distributed transaction block with ease. Therefore we raise warnings.
Relevant PG commit: cfdd03f45e6afc632fbe70519250ec19167d6765
Marco Slot
Onur Tirtir
Alexander Kukushkin
oohira
Teja Mupparti
Ahmet Gedemenli
aykut-bozkurt
Emel Şimşek
Jelte Fennema
Gokhan Gulbiz
Önder Kalacı
Hanefi Onaldi
Naisila Puka
Nitish Upreti
Sameer Awasekar
Nils Dijk
Jelte Fennema