DESCRIPTION: Fix counter that keeps track of internal depth in executor
While reviewing #3302 I ran into the `ExecutorLevel` variable which used a variable to keep the original value to restore on successful exit. I haven't explored the full space and if it is possible to get into an inconsistent state. However using `PG_TRY`/`PG_CATCH` seems generally more correct.
Given very bad things will happen if this level is not reset, I kept the failsafe of setting the variiable back to 0 on the `XactCallback` but I did add an assert to treat it as a developer bug.
* Remove unused executor codes
All of the codes of real-time executor. Some functions
in router executor still remains there because there
are common functions. We'll move them to accurate places
in the follow-up commits.
* Move GUCs to transaction mngnt and remove unused struct
* Update test output
* Get rid of references of real-time executor from code
* Warn if real-time executor is picked
* Remove lots of unused connection codes
* Removed unused code for connection restrictions
Real-time and router executors cannot handle re-using of the existing
connections within a transaction block.
Adaptive executor and COPY can re-use the connections. So, there is no
reason to keep the code around for applying the restrictions in the
placement connection logic.
Objectives:
(a) both super user and regular user should have the correct owner for the function on the worker
(b) The transactional semantics would work fine for both super user and regular user
(c) non-super-user and non-function owner would get a reasonable error message if tries to distribute the function
Co-authored-by: @serprex
When a function is marked as colocated with a distributed table,
we try delegating queries of kind "SELECT func(...)" to workers.
We currently only support this simple form, and don't delegate
forms like "SELECT f1(...), f2(...)", "SELECT f1(...) FROM ...",
or function calls inside transactions.
As a side effect, we also fix the transactional semantics of DO blocks.
Previously we didn't consider a DO block a multi-statement transaction.
Now we do.
Co-authored-by: Marco Slot <marco@citusdata.com>
Co-authored-by: serprex <serprex@users.noreply.github.com>
Co-authored-by: pykello <hadi.moshayedi@microsoft.com>
DESCRIPTION: Distribute Types to worker nodes
When to propagate
==============
There are two logical moments that types could be distributed to the worker nodes
- When they get used ( just in time distribution )
- When they get created ( proactive distribution )
The just in time distribution follows the model used by how schema's get created right before we are going to create a table in that schema, for types this would be when the table uses a type as its column.
The proactive distribution is suitable for situations where it is benificial to have the type on the worker nodes directly. They can later on be used in queries where an intermediate result gets created with a cast to this type.
Just in time creation is always the last resort, you cannot create a distributed table before the type gets created. A good example use case is; you have an existing postgres server that needs to scale out. By adding the citus extension, add some nodes to the cluster, and distribute the table. The type got created before citus existed. There was no moment where citus could have propagated the creation of a type.
Proactive is almost always a good option. Types are not resource intensive objects, there is no performance overhead of having 100's of types. If you want to use them in a query to represent an intermediate result (which happens in our test suite) they just work.
There is however a moment when proactive type distribution is not beneficial; in transactions where the type is used in a distributed table.
Lets assume the following transaction:
```sql
BEGIN;
CREATE TYPE tt1 AS (a int, b int);
CREATE TABLE t1 AS (a int PRIMARY KEY, b tt1);
SELECT create_distributed_table('t1', 'a');
\copy t1 FROM bigdata.csv
```
Types are node scoped objects; meaning the type exists once per worker. Shards however have best performance when they are created over their own connection. For the type to be visible on all connections it needs to be created and committed before we try to create the shards. Here the just in time situation is most beneficial and follows how we create schema's on the workers. Outside of a transaction block we will just use 1 connection to propagate the creation.
How propagation works
=================
Just in time
-----------
Just in time propagation hooks into the infrastructure introduced in #2882. It adds types as a supported object in `SupportedDependencyByCitus`. This will make sure that any object being distributed by citus that depends on types will now cascade into types. When types are depending them self on other objects they will get created first.
Creation later works by getting the ddl commands to create the object by its `ObjectAddress` in `GetDependencyCreateDDLCommands` which will dispatch types to `CreateTypeDDLCommandsIdempotent`.
For the correct walking of the graph we follow array types, when later asked for the ddl commands for array types we return `NIL` (empty list) which makes that the object will not be recorded as distributed, (its an internal type, dependant on the user type).
Proactive distribution
---------------------
When the user creates a type (composite or enum) we will have a hook running in `multi_ProcessUtility` after the command has been applied locally. Running after running locally makes that we already have an `ObjectAddress` for the type. This is required to mark the type as being distributed.
Keeping the type up to date
====================
For types that are recorded in `pg_dist_object` (eg. `IsObjectDistributed` returns true for the `ObjectAddress`) we will intercept the utility commands that alter the type.
- `AlterTableStmt` with `relkind` set to `OBJECT_TYPE` encapsulate changes to the fields of a composite type.
- `DropStmt` with removeType set to `OBJECT_TYPE` encapsulate `DROP TYPE`.
- `AlterEnumStmt` encapsulates changes to enum values.
Enum types can not be changed transactionally. When the execution on a worker fails a warning will be shown to the user the propagation was incomplete due to worker communication failure. An idempotent command is shown for the user to re-execute when the worker communication is fixed.
Keeping types up to date is done via the executor. Before the statement is executed locally we create a plan on how to apply it on the workers. This plan is executed after we have applied the statement locally.
All changes to types need to be done in the same transaction for types that have already been distributed and will fail with an error if parallel queries have already been executed in the same transaction. Much like foreign keys to reference tables.
/*
* local_executor.c
*
* The scope of the local execution is locally executing the queries on the
* shards. In other words, local execution does not deal with any local tables
* that are not shards on the node that the query is being executed. In that sense,
* the local executor is only triggered if the node has both the metadata and the
* shards (e.g., only Citus MX worker nodes).
*
* The goal of the local execution is to skip the unnecessary network round-trip
* happening on the node itself. Instead, identify the locally executable tasks and
* simply call PostgreSQL's planner and executor.
*
* The local executor is an extension of the adaptive executor. So, the executor uses
* adaptive executor's custom scan nodes.
*
* One thing to note that Citus MX is only supported with replication factor = 1, so
* keep that in mind while continuing the comments below.
*
* On the high level, there are 3 slightly different ways of utilizing local execution:
*
* (1) Execution of local single shard queries of a distributed table
*
* This is the simplest case. The executor kicks at the start of the adaptive
* executor, and since the query is only a single task the execution finishes
* without going to the network at all.
*
* Even if there is a transaction block (or recursively planned CTEs), as long
* as the queries hit the shards on the same, the local execution will kick in.
*
* (2) Execution of local single queries and remote multi-shard queries
*
* The rule is simple. If a transaction block starts with a local query execution,
* all the other queries in the same transaction block that touch any local shard
* have to use the local execution. Although this sounds restrictive, we prefer to
* implement in this way, otherwise we'd end-up with as complex scenarious as we
* have in the connection managements due to foreign keys.
*
* See the following example:
* BEGIN;
* -- assume that the query is executed locally
* SELECT count(*) FROM test WHERE key = 1;
*
* -- at this point, all the shards that reside on the
* -- node is executed locally one-by-one. After those finishes
* -- the remaining tasks are handled by adaptive executor
* SELECT count(*) FROM test;
*
*
* (3) Modifications of reference tables
*
* Modifications to reference tables have to be executed on all nodes. So, after the
* local execution, the adaptive executor keeps continuing the execution on the other
* nodes.
*
* Note that for read-only queries, after the local execution, there is no need to
* kick in adaptive executor.
*
* There are also few limitations/trade-offs that is worth mentioning. First, the
* local execution on multiple shards might be slow because the execution has to
* happen one task at a time (e.g., no parallelism). Second, if a transaction
* block/CTE starts with a multi-shard command, we do not use local query execution
* since local execution is sequential. Basically, we do not want to lose parallelism
* across local tasks by switching to local execution. Third, the local execution
* currently only supports queries. In other words, any utility commands like TRUNCATE,
* fails if the command is executed after a local execution inside a transaction block.
* Forth, the local execution cannot be mixed with the executors other than adaptive,
* namely task-tracker, real-time and router executors. Finally, related with the
* previous item, COPY command cannot be mixed with local execution in a transaction.
* The implication of that any part of INSERT..SELECT via coordinator cannot happen
* via the local execution.
*/
* Add tuplestore helpers
* More detailed error messages in tuplestore
* Add CreateTupleDescCopy to SetupTuplestore
* Use new SetupTuplestore helper function
* Remove unnecessary copy
* Remove comment about undefined behaviour
Before this commit, we've recorded the relation accesses in 3 different
places
- FindPlacementListConnection -- applies all executor in tx block
- StartPlacementExecutionOnSession() -- adaptive executor only
- StartPlacementListConnection() -- router/real-time only
This is different than Citus 8.2, and could lead to query execution times
increase considerably on multi-shard commands in transaction block
that are on partitioned tables.
Benchmarks:
```
1+8 c5.4xlarge cluster
Empty distributed partitioned table with 365 partitions: https://gist.github.com/onderkalaci/1edace4ed6bd6f061c8a15594865bb51#file-partitions_365-sql
./pgbench -f /tmp/multi_shard.sql -c10 -j10 -P 1 -T 120 postgres://citus:w3r6KLJpv3mxe9E-NIUeJw@c.fy5fkjcv45vcepaogqcaskmmkee.db.citusdata.com:5432/citus?sslmode=require
cat /tmp/multi_shard.sql
BEGIN;
DELETE FROM collections_list;
DELETE FROM collections_list;
DELETE FROM collections_list;
COMMIT;
cat /tmp/single_shard.sql
BEGIN;
DELETE FROM collections_list WHERE key = :aid;
DELETE FROM collections_list WHERE key = :aid;
DELETE FROM collections_list WHERE key = :aid;
COMMIT;
cat /tmp/mix.sql
BEGIN;
DELETE FROM collections_list WHERE key = :aid;
DELETE FROM collections_list WHERE key = :aid;
DELETE FROM collections_list WHERE key = :aid;
DELETE FROM collections_list;
DELETE FROM collections_list;
DELETE FROM collections_list;
COMMIT;
```
The table shows `latency average` of pgbench runs explained above, so we have a pretty solid improvement even over 8.2.2.
| Test | Citus 8.2.2 | Citus 8.3.1 | Citus 8.3.2 (this branch) | Citus 8.3.1 (FKEYs disabled via GUC) |
| ------------- | ------------- | ------------- |------------- | ------------- |
|multi_shard | 2370.083 ms |3605.040 ms |1324.094 ms |1247.255 ms |
| single_shard | 85.338 ms |120.934 ms |73.216 ms | 78.765 ms |
| mix | 2434.459 ms | 3727.080 ms |1306.456 ms | 1280.326 ms |
With this commit, we're introducing the Adaptive Executor.
The commit message consists of two distinct sections. The first part explains
how the executor works. The second part consists of the commit messages of
the individual smaller commits that resulted in this commit. The readers
can search for the each of the smaller commit messages on
https://github.com/citusdata/citus and can learn more about the history
of the change.
/*-------------------------------------------------------------------------
*
* adaptive_executor.c
*
* The adaptive executor executes a list of tasks (queries on shards) over
* a connection pool per worker node. The results of the queries, if any,
* are written to a tuple store.
*
* The concepts in the executor are modelled in a set of structs:
*
* - DistributedExecution:
* Execution of a Task list over a set of WorkerPools.
* - WorkerPool
* Pool of WorkerSessions for the same worker which opportunistically
* executes "unassigned" tasks from a queue.
* - WorkerSession:
* Connection to a worker that is used to execute "assigned" tasks
* from a queue and may execute unasssigned tasks from the WorkerPool.
* - ShardCommandExecution:
* Execution of a Task across a list of placements.
* - TaskPlacementExecution:
* Execution of a Task on a specific placement.
* Used in the WorkerPool and WorkerSession queues.
*
* Every connection pool (WorkerPool) and every connection (WorkerSession)
* have a queue of tasks that are ready to execute (readyTaskQueue) and a
* queue/set of pending tasks that may become ready later in the execution
* (pendingTaskQueue). The tasks are wrapped in a ShardCommandExecution,
* which keeps track of the state of execution and is referenced from a
* TaskPlacementExecution, which is the data structure that is actually
* added to the queues and describes the state of the execution of a task
* on a particular worker node.
*
* When the task list is part of a bigger distributed transaction, the
* shards that are accessed or modified by the task may have already been
* accessed earlier in the transaction. We need to make sure we use the
* same connection since it may hold relevant locks or have uncommitted
* writes. In that case we "assign" the task to a connection by adding
* it to the task queue of specific connection (in
* AssignTasksToConnections). Otherwise we consider the task unassigned
* and add it to the task queue of a worker pool, which means that it
* can be executed over any connection in the pool.
*
* A task may be executed on multiple placements in case of a reference
* table or a replicated distributed table. Depending on the type of
* task, it may not be ready to be executed on a worker node immediately.
* For instance, INSERTs on a reference table are executed serially across
* placements to avoid deadlocks when concurrent INSERTs take conflicting
* locks. At the beginning, only the "first" placement is ready to execute
* and therefore added to the readyTaskQueue in the pool or connection.
* The remaining placements are added to the pendingTaskQueue. Once
* execution on the first placement is done the second placement moves
* from pendingTaskQueue to readyTaskQueue. The same approach is used to
* fail over read-only tasks to another placement.
*
* Once all the tasks are added to a queue, the main loop in
* RunDistributedExecution repeatedly does the following:
*
* For each pool:
* - ManageWorkPool evaluates whether to open additional connections
* based on the number unassigned tasks that are ready to execute
* and the targetPoolSize of the execution.
*
* Poll all connections:
* - We use a WaitEventSet that contains all (non-failed) connections
* and is rebuilt whenever the set of active connections or any of
* their wait flags change.
*
* We almost always check for WL_SOCKET_READABLE because a session
* can emit notices at any time during execution, but it will only
* wake up WaitEventSetWait when there are actual bytes to read.
*
* We check for WL_SOCKET_WRITEABLE just after sending bytes in case
* there is not enough space in the TCP buffer. Since a socket is
* almost always writable we also use WL_SOCKET_WRITEABLE as a
* mechanism to wake up WaitEventSetWait for non-I/O events, e.g.
* when a task moves from pending to ready.
*
* For each connection that is ready:
* - ConnectionStateMachine handles connection establishment and failure
* as well as command execution via TransactionStateMachine.
*
* When a connection is ready to execute a new task, it first checks its
* own readyTaskQueue and otherwise takes a task from the worker pool's
* readyTaskQueue (on a first-come-first-serve basis).
*
* In cases where the tasks finish quickly (e.g. <1ms), a single
* connection will often be sufficient to finish all tasks. It is
* therefore not necessary that all connections are established
* successfully or open a transaction (which may be blocked by an
* intermediate pgbouncer in transaction pooling mode). It is therefore
* essential that we take a task from the queue only after opening a
* transaction block.
*
* When a command on a worker finishes or the connection is lost, we call
* PlacementExecutionDone, which then updates the state of the task
* based on whether we need to run it on other placements. When a
* connection fails or all connections to a worker fail, we also call
* PlacementExecutionDone for all queued tasks to try the next placement
* and, if necessary, mark shard placements as inactive. If a task fails
* to execute on all placements, the execution fails and the distributed
* transaction rolls back.
*
* For multi-row INSERTs, tasks are executed sequentially by
* SequentialRunDistributedExecution instead of in parallel, which allows
* a high degree of concurrency without high risk of deadlocks.
* Conversely, multi-row UPDATE/DELETE/DDL commands take aggressive locks
* which forbids concurrency, but allows parallelism without high risk
* of deadlocks. Note that this is unrelated to SEQUENTIAL_CONNECTION,
* which indicates that we should use at most one connection per node, but
* can run tasks in parallel across nodes. This is used when there are
* writes to a reference table that has foreign keys from a distributed
* table.
*
* Execution finishes when all tasks are done, the query errors out, or
* the user cancels the query.
*
*-------------------------------------------------------------------------
*/
All the commits involved here:
* Initial unified executor prototype
* Latest changes
* Fix rebase conflicts to master branch
* Add missing variable for assertion
* Ensure that master_modify_multiple_shards() returns the affectedTupleCount
* Adjust intermediate result sizes
The real-time executor uses COPY command to get the results
from the worker nodes. Unified executor avoids that which
results in less data transfer. Simply adjust the tests to lower
sizes.
* Force one connection per placement (or co-located placements) when requested
The existing executors (real-time and router) always open 1 connection per
placement when parallel execution is requested.
That might be useful under certain circumstances:
(a) User wants to utilize as much as CPUs on the workers per
distributed query
(b) User has a transaction block which involves COPY command
Also, lots of regression tests rely on this execution semantics.
So, we'd enable few of the tests with this change as well.
* For parameters to be resolved before using them
For the details, see PostgreSQL's copyParamList()
* Unified executor sorts the returning output
* Ensure that unified executor doesn't ignore sequential execution of DDLJob's
Certain DDL commands, mainly creating foreign keys to reference tables,
should be executed sequentially. Otherwise, we'd end up with a self
distributed deadlock.
To overcome this situaiton, we set a flag `DDLJob->executeSequentially`
and execute it sequentially. Note that we have to do this because
the command might not be called within a transaction block, and
we cannot call `SetLocalMultiShardModifyModeToSequential()`.
This fixes at least two test: multi_insert_select_on_conflit.sql and
multi_foreign_key.sql
Also, I wouldn't mind scattering local `targetPoolSize` variables within
the code. The reason is that we'll soon have a GUC (or a global
variable based on a GUC) that'd set the pool size. In that case, we'd
simply replace `targetPoolSize` with the global variables.
* Fix 2PC conditions for DDL tasks
* Improve closing connections that are not fully established in unified execution
* Support foreign keys to reference tables in unified executor
The idea for supporting foreign keys to reference tables is simple:
Keep track of the relation accesses within a transaction block.
- If a parallel access happens on a distributed table which
has a foreign key to a reference table, one cannot modify
the reference table in the same transaction. Otherwise,
we're very likely to end-up with a self-distributed deadlock.
- If an access to a reference table happens, and then a parallel
access to a distributed table (which has a fkey to the reference
table) happens, we switch to sequential mode.
Unified executor misses the function calls that marks the relation
accesses during the execution. Thus, simply add the necessary calls
and let the logic kick in.
* Make sure to close the failed connections after the execution
* Improve comments
* Fix savepoints in unified executor.
* Rebuild the WaitEventSet only when necessary
* Unclaim connections on all errors.
* Improve failure handling for unified executor
- Implement the notion of errorOnAnyFailure. This is similar to
Critical Connections that the connection managament APIs provide
- If the nodes inside a modifying transaction expand, activate 2PC
- Fix few bugs related to wait event sets
- Mark placement INACTIVE during the execution as much as possible
as opposed to we do in the COMMIT handler
- Fix few bugs related to scheduling next placement executions
- Improve decision on when to use 2PC
Improve the logic to start a transaction block for distributed transactions
- Make sure that only reference table modifications are always
executed with distributed transactions
- Make sure that stored procedures and functions are executed
with distributed transactions
* Move waitEventSet to DistributedExecution
This could also be local to RunDistributedExecution(), but in that case
we had to mark it as "volatile" to avoid PG_TRY()/PG_CATCH() issues, and
cast it to non-volatile when doing WaitEventSetFree(). We thought that
would make code a bit harder to read than making this non-local, so we
move it here. See comments for PG_TRY() in postgres/src/include/elog.h
and "man 3 siglongjmp" for more context.
* Fix multi_insert_select test outputs
Two things:
1) One complex transaction block is now supported. Simply update
the test output
2) Due to dynamic nature of the unified executor, the orders of
the errors coming from the shards might change (e.g., all of
the queries on the shards would fail, but which one appears
on the error message?). To fix that, we simply added it to
our shardId normalization tool which happens just before diff.
* Fix subeury_and_cte test
The error message is updated from:
failed to execute task
To:
more than one row returned by a subquery or an expression
which is a lot clearer to the user.
* Fix intermediate_results test outputs
Simply update the error message from:
could not receive query results
to
result "squares" does not exist
which makes a lot more sense.
* Fix multi_function_in_join test
The error messages update from:
Failed to execute task XXX
To:
function f(..) does not exist
* Fix multi_query_directory_cleanup test
The unified executor does not create any intermediate files.
* Fix with_transactions test
A test case that just started to work fine
* Fix multi_router_planner test outputs
The error message is update from:
Could not receive query results
To:
Relation does not exists
which is a lot more clearer for the users
* Fix multi_router_planner_fast_path test
The error message is update from:
Could not receive query results
To:
Relation does not exists
which is a lot more clearer for the users
* Fix isolation_copy_placement_vs_modification by disabling select_opens_transaction_block
* Fix ordering in isolation_multi_shard_modify_vs_all
* Add executor locks to unified executor
* Make sure to allocate enought WaitEvents
The previous code was missing the waitEvents for the latch and
postmaster death.
* Fix rebase conflicts for master rebase
* Make sure that TRUNCATE relies on unified executor
* Implement true sequential execution for multi-row INSERTS
Execute the individual tasks executed one by one. Note that this is different than
MultiShardConnectionType == SEQUENTIAL_CONNECTION case (e.g., sequential execution
mode). In that case, running the tasks across the nodes in parallel is acceptable
and implemented in that way.
However, the executions that are qualified here would perform poorly if the
tasks across the workers are executed in parallel. We currently qualify only
one class of distributed queries here, multi-row INSERTs. If we do not enforce
true sequential execution, concurrent multi-row upserts could easily form
a distributed deadlock when the upserts touch the same rows.
* Remove SESSION_LIFESPAN flag in unified_executor
* Apply failure test updates
We've changed the failure behaviour a bit, and also the error messages
that show up to the user. This PR covers majority of the updates.
* Unified executor honors citus.node_connection_timeout
With this commit, unified executor errors out if even
a single connection cannot be established within
citus.node_connection_timeout.
And, as a side effect this fixes failure_connection_establishment
test.
* Properly increment/decrement pool size variables
Before this commit, the idle and active connection
counts were not properly calculated.
* insert_select_executor goes through unified executor.
* Add missing file for task tracker
* Modify ExecuteTaskListExtended()'s signature
* Sort output of INSERT ... SELECT ... RETURNING
* Take partition locks correctly in unified executor
* Alternative implementation for force_max_query_parallelization
* Fix compile warnings in unified executor
* Fix style issues
* Decrement idleConnectionCount when idle connection is lost
* Always rebuild the wait event sets
In the previous implementation, on waitFlag changes, we were only
modifying the wait events. However, we've realized that it might
be an over optimization since (a) we couldn't see any performance
benefits (b) we see some errors on failures and because of (a)
we prefer to disable it now.
* Make sure to allocate enough sized waitEventSet
With multi-row INSERTs, we might have more sessions than
task*workerCount after few calls of RunDistributedExecution()
because the previous sessions would also be alive.
Instead, re-allocate events when the connectino set changes.
* Implement SELECT FOR UPDATE on reference tables
On master branch, we do two extra things on SELECT FOR UPDATE
queries on reference tables:
- Acquire executor locks
- Execute the query on all replicas
With this commit, we're implementing the same logic on the
new executor.
* SELECT FOR UPDATE opens transaction block even if SelectOpensTransactionBlock disabled
Otherwise, users would be very confused and their logic is very likely
to break.
* Fix build error
* Fix the newConnectionCount calculation in ManageWorkerPool
* Fix rebase conflicts
* Fix minor test output differences
* Fix citus indent
* Remove duplicate sorts that is added with rebase
* Create distributed table via executor
* Fix wait flags in CheckConnectionReady
* failure_savepoints output for unified executor.
* failure_vacuum output (pg 10) for unified executor.
* Fix WaitEventSetWait timeout in unified executor
* Stabilize failure_truncate test output
* Add an ORDER BY to multi_upsert
* Fix regression test outputs after rebase to master
* Add executor.c comment
* Rename executor.c to adaptive_executor.c
* Do not schedule tasks if the failed placement is not ready to execute
Before the commit, we were blindly scheduling the next placement executions
even if the failed placement is not on the ready queue. Now, we're ensuring
that if failed placement execution is on a failed pool or session where the
execution is on the pendingQueue, we do not schedule the next task. Because
the other placement execution should be already running.
* Implement a proper custom scan node for adaptive executor
- Switch between the executors, add GUC to set the pool size
- Add non-adaptive regression test suites
- Enable CIRCLE CI for non-adaptive tests
- Adjust test output files
* Add slow start interval to the executor
* Expose max_cached_connection_per_worker to user
* Do not start slow when there are cached connections
* Consider ExecutorSlowStartInterval in NextEventTimeout
* Fix memory issues with ReceiveResults().
* Disable executor via TaskExecutorType
* Make sure to execute the tests with the other executor
* Use task_executor_type to enable-disable adaptive executor
* Remove useless code
* Adjust the regression tests
* Add slow start regression test
* Rebase to master
* Fix test failures in adaptive executor.
* Rebase to master - 2
* Improve comments & debug messages
* Set force_max_query_parallelization in isolation_citus_dist_activity
* Force max parallelization for creating shards when asked to use exclusive connection.
* Adjust the default pool size
* Expand description of max_adaptive_executor_pool_size GUC
* Update warnings in FinishRemoteTransactionCommit()
* Improve session clean up at the end of execution
Explicitly list all the states that the execution might end,
otherwise warn.
* Remove MULTI_CONNECTION_WAIT_RETRY which is not used at all
* Add more ORDER BYs to multi_mx_partitioning
Adds support for propagation of SET LOCAL commands to all workers
involved in a query. For now, SET SESSION (i.e. plain SET) is not
supported whatsoever, though this code is intended as somewhat of a
base for implementing such support in the future.
As SET LOCAL modifications are scoped to the body of a BEGIN/END xact
block, queries wishing to use SET LOCAL propagation must be within such
a block. In addition, subsequent modifications after e.g. any SAVEPOINT
or ROLLBACK statements will correspondingly push or pop variable mod-
ifications onto an internal stack such that the behavior of changed
values across the cluster will be identical to such behavior on e.g.
single-node PostgreSQL (or equivalently, what values are visible to
the end user by running SHOW on such variables on the coordinator).
If nodes enter the set of participants at some point after SET LOCAL
modifications (or SAVEPOINT, ROLLBACK, etc.) have occurred, the SET
variable state is eagerly propagated to them upon their entrance (this
is identical to, and indeed just augments, the existing logic for the
propagation of the SAVEPOINT "stack").
A new GUC (citus.propagate_set_commands) has been added to control this
behavior. Though the code suggests the valid settings are 'none', 'local',
'session', and 'all', only 'none' (the default) and 'local' are presently
implemented: attempting to use other values will result in an error.
Instead of scattering the code around, we move all the
logic into a single function.
This will help supporting foreign keys to reference tables
in the unified executor with a single line of change, just
calling this function.
Before this commit, round-robin task assignment policy was relying
on the taskId. Thus, even inside a transaction, the tasks were
assigned to different nodes. This was especially problematic
while reading from reference tables within transaction blocks.
Because, we had to expand the distributed transaction to many
nodes that are not necessarily already in the distributed transaction.
We had recently fixed a spinlock issue due to functions
failing, but spinlock is not being released.
This is the continuation of that work to eliminate possible
regression of the issue. Function calls that are moved out of
spinlock scope are macros and plain type casting. However,
depending on the configuration they have an alternate implementation
in PG source that performs memory allocation.
This commit moves last bit of codes to out of spinlock for completion purposes.
A spinlock is not released when an exception is thrown after
spinlock is acquired. This has caused infinite wait and eventual
crash in maintenance daemon.
This work moves the code than can fail to the outside of spinlock
scope so that in the case of failure spinlock is not left locked
since it was not locked in the first place.
Postgresql loads shared libraries before calculating MaxBackends.
However, Citus relies on MaxBackends being set. Thus, with this
commit we use the same steps to calculate MaxBackends while
Citus is being loaded (e.g., PG_Init is called).
Note that this is safe since all the elements that are used to
calculate MaxBackends are PGC_POSTMASTER gucs and a constant
value.
In case a failure happens when a transaction is failed on PREPARE,
we used to hit an assertion for ensuring there is no pending
activity on the connection. However, that's not true after the
changes in #2031. Thus, we've replaced the assertion with a more
generic function call to consume any pending activity, if exists.
In the distributed deadlock detection design, we concluded that prepared transactions
cannot be part of a distributed deadlock. The idea is that (a) when the transaction
is prepared it already acquires all the locks, so cannot be part of a deadlock
(b) even if some other processes blocked on the prepared transaction, prepared transactions
would eventually be committed (or rollbacked) and the system will continue operating.
With the above in mind, we probably had a mistake in terms of memory allocations. For each
backend initialized, we keep a `BackendData` struct. The bug we've introduced is that, we
assumed there would only be `MaxBackend` number of backends. However, `MaxBackends` doesn't
include the prepared transactions and axuliary processes. When you check Postgres' InitProcGlobal`
you'd see that `TotalProcs = MaxBackends + NUM_AUXILIARY_PROCS + max_prepared_xacts;`
This commit aligns with total procs processed with that.
With this commit, we implement two views that are very similar
to pg_stat_activity, but showing queries that are involved in
distributed queries:
- citus_dist_stat_activity: Shows all the distributed queries
- citus_worker_stat_activity: Shows all the queries on the shards
that are initiated by distributed queries.
Both views have the same columns in the outputs. In very basic terms, both of the views
are meant to provide some useful insights about the distributed
transactions within the cluster. As the names reveal, both views are similar to pg_stat_activity.
Also note that these views can be pretty useful on Citus MX clusters.
Note that when the views are queried from the worker nodes, they'd not show the distributed
transactions that are initiated from the coordinator node. The reason is that the worker
nodes do not know the host/port of the coordinator. Thus, it is advisable to query the
views from the coordinator.
If we bucket the columns that the views returns, we'd end up with the following:
- Hostnames and ports:
- query_hostname, query_hostport: The node that the query is running
- master_query_host_name, master_query_host_port: The node in the cluster
initiated the query.
Note that for citus_dist_stat_activity view, the query_hostname-query_hostport
is always the same with master_query_host_name-master_query_host_port. The
distinction is mostly relevant for citus_worker_stat_activity. For example,
on Citus MX, a users starts a transaction on Node-A, which starts worker
transactions on Node-B and Node-C. In that case, the query hostnames would be
Node-B and Node-C whereas the master_query_host_name would Node-A.
- Distributed transaction related things:
This is mostly the process_id, distributed transactionId and distributed transaction
number.
- pg_stat_activity columns:
These two views get all the columns from pg_stat_activity. We're basically joining
pg_stat_activity with get_all_active_transactions on process_id.
We previously implemented OTHER_WORKERS_WITH_METADATA tag. However,
that was wrong. See the related discussion:
https://github.com/citusdata/citus/issues/2320
Instead, we switched using OTHER_WORKER_NODES and make the command
that we're running optional such that even if the node is not a
metadata node, we won't be in trouble.
This commit enables support for TRUNCATE on both
distributed table and reference tables.
The basic idea is to acquire lock on the relation by sending
the TRUNCATE command to all metedata worker nodes. We only
skip sending the TRUNCATE command to the node that actually
executus the command to prevent a self-distributed-deadlock.
When a hash distributed table have a foreign key to a reference
table, there are few restrictions we have to apply in order to
prevent distributed deadlocks or reading wrong results.
The necessity to apply the restrictions arise from cascading
nature of foreign keys. When a foreign key on a reference table
cascades to a distributed table, a single operation over a single
connection can acquire locks on multiple shards of the distributed
table. Thus, any parallel operation on that distributed table, in the
same transaction should not open parallel connections to the shards.
Otherwise, we'd either end-up with a self-distributed deadlock or
read wrong results.
As briefly described above, the restrictions that we apply is done
by tracking the distributed/reference relation accesses inside
transaction blocks, and act accordingly when necessary.
The two main rules are as follows:
- Whenever a parallel distributed relation access conflicts
with a consecutive reference relation access, Citus errors
out
- Whenever a reference relation access is followed by a
conflicting parallel relation access, the execution mode
is switched to sequential mode.
There are also some other notes to mention:
- If the user does SET LOCAL citus.multi_shard_modify_mode
TO 'sequential';, all the queries should simply work with
using one connection per worker and sequentially executing
the commands. That's obviously a slower approach than Citus'
usual parallel execution. However, we've at least have a way
to run all commands successfully.
- If an unrelated parallel query executed on any distributed
table, we cannot switch to sequential mode. Because, the essense
of sequential mode is using one connection per worker. However,
in the presence of a parallel connection, the connection manager
picks those connections to execute the commands. That contradicts
with our purpose, thus we error out.
- COPY to a distributed table cannot be executed in sequential mode.
Thus, if we switch to sequential mode and COPY is executed, the
operation fails and there is currently no way of implementing that.
Note that, when the local table is not empty and create_distributed_table
is used, citus uses COPY internally. Thus, in those cases,
create_distributed_table() will also fail.
- There is a GUC called citus.enforce_foreign_key_restrictions
to disable all the checks. We added that GUC since the restrictions
we apply is sometimes a bit more restrictive than its necessary.
The user might want to relax those. Similarly, if you don't have
CASCADEing reference tables, you might consider disabling all the
checks.
Errors thrown in the COMMIT handler will cause Postgres to segfault,
there's nothing it can do it abort the transaction by the time that
handler is called!
RemoveIntermediateResultsDirectory is problematic for two reasons:
- It has calls to ereport(ERROR which have been known to trigger
- It makes memory allocations which raise ERRORs when they fail
Once the COMMIT process has begun we don't use the intermediate results,
so it's safe to remove them a little earlier in the process. A failure
here will abort the transaction. That's pretty unnecessary, it's not
that important that we remove the results, but it's still better than a
crash.
In case a failure happens when a transaction is rollbacked,
we used to hit an assertion for ensuring there is no pending
activity on the connection. However, that's not true after the
changes in #2031. Thus, we've replaced the assertion with a more
generic function call to consume any pending activity, if exists.
- changes in ruleutils_11.c is reflected
- vacuum statement api change is handled. We now allow
multi-table vacuum commands.
- some other function header changes are reflected
- api conflicts between PG11 and earlier versions
are handled by adding shims in version_compat.h
- various regression tests are fixed due output and
functionality in PG1
- no change is made to support new features in PG11
they need to be handled by new commit
VLAs aren't supported by Visual Studio.
- Remove all existing instances of VLAs.
- Add a flag, -Werror=vla, which makes gcc refuse to compile if we add
VLAs in the future.
- variable length arrays (VLAs) do not work with Visual Studio
- fix an off-by-one error. We incorrectly assumed there would always at
least as many edges as there were nodes.
- refactor: reduce scope of transactionNodeStack by moving it into the
function which uses it.
- refactor: break up the distinct uses of currentStackDepth into
separate variables.
It's against our coding convention to call functions inside parameter
lists; when single-stepping with a debugger it's difficult to determine
what the function returned.
That wouldn't be good enough reason to change this code but while
porting Citus to Windows I ran into this line of code.
assign_distributed_transaction_id was called with a weird timestamp and
I wasn't able to find the problem without first making this change.
With this fix, we traverse the graph with DFS which was originally
intended. Note that, before the fix, we traverse the graph with BFS
which might lead to killing some unrelated backend that is not
involved in the distributed deadlock.
This commit introduces a new GUC to limit the intermediate
result size which we handle when we use read_intermediate_result
function for CTEs and complex subqueries.
Postgres provides OS agnosting formatting macros for
formatting 64 bit numbers. Replaced %ld %lu with
INT64_FORMAT and UINT64_FORMAT respectively.
Also found some incorrect usages of formatting
flags and fixed them.
Autovacuum process cancels itself if any modification starts
on the table in order to avoid blocking your regular Postgres
sessions. That's normal and expected. Thus, any locks held by
autovacuum process cannot involve in a distributed deadlock
since it'll be released if needed.
These locks are held for a very short duration time and cannot
contribute to a deadlock. Speculative locks are used by Postgres
for internal notification mechanism among transactions.
Short-term share/exclusive page-level locks are used for
read/write access. Locks are released immediately after
each index row is fetched or inserted.
Since those locks may not lead to any deadlocks, it's safe
to ignore them in the distributed deadlock detection.
We should skip if the process blocked on the relation
extension since those locks are hold for a short duration
while the relation is actually extended on the disk and
released as soon as the extension is done. Thus, recording
such waits on our lock graphs could yield detecting wrong
distributed deadlocks.
We sent multiple commands to worker when starting a transaction.
Previously we only checked the result of the first command that
is transaction 'BEGIN' which always succeeds. Any failure on
following commands were not checked.
With this commit, we make sure all command results are checked.
If there is any error we report the first error found.
If after the distributed deadlock detection decides to cancel
a backend, the backend has been terminated/killed/cancelled
externally, we might be accessing to a NULL pointer. This commit
prevents that case by ignoring the current distributed deadlock.
This change adds support for SAVEPOINT, ROLLBACK TO SAVEPOINT, and RELEASE SAVEPOINT.
When transaction connections are not established yet, savepoints are kept in a stack and sent to the worker when the connection is later established. After establishing connections, savepoint commands are sent as they arrive.
This change fixes#1493 .
We added a new field to the transaction id that is set to true only
for the transactions initialized on the coordinator. This is only
useful for MX in order to distinguish the transaction that started
the distributed transaction on the coordinator where we could
have the same transactions' worker queries on the same node.
We added a new GUC citus.log_distributed_deadlock_detection
which is off by default. When set to on, we log some debug messages
related to the distributed deadlock to the server logs.
This commit adds all the necessary pieces to do the distributed
deadlock detection.
Each distributed transaction is already assigned with distributed
transaction ids introduced with
3369f3486f. The dependency among the
distributed transactions are gathered with
80ea233ec1.
With this commit, we implement a DFS (depth first seach) on the
dependency graph and search for cycles. Finding a cycle reveals
a distributed deadlock.
Once we find the deadlock, we examine the path that the cycle exists
and cancel the youngest distributed transaction.
Note that, we're not yet enabling the deadlock detection by default
with this commit.
This GUC has two settings, 'always' and 'never'. When it's set to
'never' all behavior stays exactly as it was prior to this commit. When
it's set to 'always' only SELECT queries are allowed to run, and only
secondary nodes are used when processing those queries.
Add some helper functions:
- WorkerNodeIsSecondary(), checks the noderole of the worker node
- WorkerNodeIsReadable(), returns whether we're currently allowed to
read from this node
- ActiveReadableNodeList(), some functions (namely, the ones on the
SELECT path) don't require working with Primary Nodes. They should call
this function instead of ActivePrimaryNodeList(), because the latter
will error out in contexts where we're not allowed to write to nodes.
- ActiveReadableNodeCount(), like the above, replaces
ActivePrimaryNodeCount().
- EnsureModificationsCanRun(), error out if we're not currently allowed
to run queries which modify data. (Either we're in read-only mode or
use_secondary_nodes is set)
Some parts of the code were switched over to use readable nodes instead
of primary nodes:
- Deadlock detection
- DistributedTableSize,
- the router, real-time, and task tracker executors
- ShardPlacement resolution
We use the backend shared memory lock for preventing
new backends to be part of a new distributed transaction
or an existing backend to leave a distributed transaction
while we're reading the all backends' data.
The primary goal is to provide consistent view of the
current distributed transactions while doing the
deadlock detection.
In this commit, we add ability to convert global wait edges
into adjacency list with the following format:
[transactionId] = [transactionNode->waitsFor {list of waiting transaction nodes}]
- master_add_node enforces that there is only one primary per group
- there's also a trigger on pg_dist_node to prevent multiple primaries
per group
- functions in metadata cache only return primary nodes
- Rename ActiveWorkerNodeList -> ActivePrimaryNodeList
- Rename WorkerGetLive{Node->Group}Count()
- Refactor WorkerGetRandomCandidateNode
- master_remove_node only complains about active shard placements if the
node being removed is a primary.
- master_remove_node only deletes all reference table placements in the
group if the node being removed is the primary.
- Rename {Node->NodeGroup}HasShardPlacements, this reflects the behavior it
already had.
- Rename DeleteAllReferenceTablePlacementsFrom{Node->NodeGroup}. This also
reflects the behavior it already had, but the new signature forces the
caller to pass in a groupId
- Rename {WorkerGetLiveGroup->ActivePrimaryNode}Count
This commit adds distributed transaction id infrastructure in
the scope of distributed deadlock detection.
In general, the distributed transaction id consists of a tuple
in the form of: `(databaseId, initiatorNodeIdentifier, transactionId,
timestamp)`.
Briefly, we add a shared memory block on each node, which holds some
information per backend (i.e., an array `BackendData backends[MaxBackends]`).
Later, on each coordinated transaction, Citus sends
`SELECT assign_distributed_transaction_id()` right after `BEGIN`.
For that backend on the worker, the distributed transaction id is set to
the values assigned via the function call.
The aim of the above is to correlate the transactions on the coordinator
to the transactions on the worker nodes.
Comes with a few changes:
- Change the signature of some functions to accept groupid
- InsertShardPlacementRow
- DeleteShardPlacementRow
- UpdateShardPlacementState
- NodeHasActiveShardPlacements returns true if the group the node is a
part of has any active shard placements
- TupleToShardPlacement now returns ShardPlacements which have NULL
nodeName and nodePort.
- Populate (nodeName, nodePort) when creating ShardPlacements
- Disallow removing a node if it contains any shard placements
- DeleteAllReferenceTablePlacementsFromNode matches based on group. This
doesn't change behavior for now (while there is only one node per
group), but means in the future callers should be careful about
calling it on a secondary node, it'll delete placements on the primary.
- Create concept of a GroupShardPlacement, which represents an actual
tuple in pg_dist_placement and is distinct from a ShardPlacement,
which has been resolved to a specific node. In the future
ShardPlacement should be renamed to NodeShardPlacement.
- Create some triggers which allow existing code to continue to insert
into and update pg_dist_shard_placement as if it still existed.
Uncrustify 0.65 appears to have changed some defaults, resulting in
breakages for those of us who have already upgraded; Travis still uses
Uncrustify 0.64, but these changes work with both versions (assuming
appropriately updated config), so this should permit use of either
version for the time being.
Previously we used ForgetResults() in StartRemoteTransactionAbort() -
that's problematic because there might still be an ongoing statement,
and this causes us to wait for its completion. That e.g. happens when
a statement running on the coordinator is cancelled.
Adds support for PostgreSQL 10 by copying in the requisite ruleutils
and updating all API usages to conform with changes in PostgreSQL 10.
Most changes are fairly minor but they are numerous. One particular
obstacle was the change in \d behavior in PostgreSQL 10's psql; I had
to add SQL implementations (views, mostly) to mimic the pre-10 output.
With this change we add an option to add a node without replicating all reference
tables to that node. If a node is added with this option, we mark the node as
inactive and no queries will sent to that node.
We also added two new UDFs;
- master_activate_node(host, port):
- marks node as active and replicates all reference tables to that node
- master_add_inactive_node(host, port):
- only adds node to pg_dist_node
- Break CheckShardPlacements into multiple functions (The most important
is MarkFailedShardPlacements), so that we can get rid of the global
CoordinatedTransactionUses2PC.
- Call MarkFailedShardPlacements in the router executor, so we mark
shards as invalid and stop using them while inside transaction blocks.
If any placements fail it doesn't update shard statistics on those placements.
A minor enabling refactor: Make CoordinatedTransactionUses2PC public (it used to be CoordinatedTransactionUse2PC but that symbol already existed, so renamed it as well)
This enables proper transactional behaviour for copy and relaxes some
restrictions like combining COPY with single-row modifications. It
also provides the basis for relaxing restrictions further, and for
optionally allowing connection caching.
Nils Dijk
Marco Slot
SaitTalhaNisanci
Hadi Moshayedi
Hadi Moshayedi
Philip Dubé
Jelte Fennema
Önder Kalacı
Marco Slot
Jason Petersen
Nils Dijk
Jason Petersen
Murat Tuncer
Dimitri Fontaine
Jason Petersen
velioglu
mehmet furkan şahin
Brian Cloutier
Matthew Wozniczka
Murat Tuncer
Andres Freund
Hadi Moshayedi
Burak Yucesoy
Brian Cloutier