DESCRIPTION: Drop `SHARD_STATE_TO_DELETE` and use the cleanup records
instead
Drops the shard state that is used to mark shards as orphaned. Now we
insert cleanup records into `pg_dist_cleanup` so "orphaned" shards will
be dropped either by maintenance daemon or internal cleanup calls. With
this PR, we make the "cleanup orphaned shards" functions to be no-op, as
they would not be needed anymore.
This PR includes some naming changes about placement functions. We don't
need functions that filter orphaned shards, as there will be no orphaned
shards anymore.
We will also be introducing a small script with this PR, for users with
orphaned shards. We'll basically delete the orphaned shard entries from
`pg_dist_placement` and insert cleanup records into `pg_dist_cleanup`
for each one of them, during Citus upgrade.
We also have a lot of flakiness fixes in this PR.
Co-authored-by: Jelte Fennema <github-tech@jeltef.nl>
Sometimes our `isolation_insert_vs_vacuum` test would fail like this.
```diff
step s2-vacuum-analyze:
VACUUM ANALYZE test_insert_vacuum;
-
+ <waiting ...>
step s1-commit:
COMMIT;
+step s2-vacuum-analyze: <... completed>
```
The reason seems to be that VACUUM ANALYZE tries to take some locks that
conflict with the other transaction, but these locks somehow get
released or VACUUM ANALYZE stops waiting for them. This is somewhat
expected since VACUUM has some special locking logic.
To solve the flakyness we now trigger VACUUM ANALYZE to always report as
blocking and after that we wait explicitly wait for it to complete. This
is done
like is suggested by the flaky test tips from postgres:
c68a183990/src/test/isolation/README (L152)
I've confirmed that this fixes the issue suing our flaky-test-debugging
CI workflow.
DESCRIPTION: Defers cleanup after a failure in shard move or split
We don't need to do a cleanup in case of failure on a shard transfer or
split anymore. Because,
* Maintenance daemon will clean them up anyway.
* We trigger a cleanup at the beginning of shard transfers/splits.
* The cleanup on failure logic also can fail sometimes and instead of
the original error, we throw the error that is raised by the cleanup
procedure, and it causes confusion.
* Skip some exceptional test files in the flaky workflow, like
multi_extension
* Run some tests without a schedule, like single_node_enterprise
* Use minimal schedule for the tests in split and operations schedules
DESCRIPTION: Cleanup the shard on the target node in case of a
failed/aborted shard move
Inserts a cleanup record for the moved shard placement on the target
node. If the move operation succeeds, the record will be deleted. If
not, it will remain there to be cleaned up later.
fixes: #6580
* Drop enterprise_split_schedule as it's not even called in our CI
pipeline. It's actually a subset of split_schedule, except for
`citus_split_shard_by_split_points_deferred_drop`. Added that one into
split_schedule and dropped the enterprise one.
* Delete `citus_non_blocking_shard_split_cleanup.out`, as there is no
sql file for it. It seems it's renamed to some other test and the sql
file is deleted, but we forgot to delete the output file.
* 6 test files are chained to each other with dependent objects. Unified
them into one test file so that the flaky check will not fail for them
anymore.
* Some cleanup lines to prevent the flakiness check from failing.
We have several version checks in our Citus upgrade tests. However, as
we drop support for PG versions, we need to update the Citus versions
used in our CI images. Therefore we must compare Citus versions in our
tests instead of using equality checks so that the queries are ran in
all the associated Citus versions.
For example, we have many conditionals where we early exit if the Citus
version is not equal to 9.0. However, as of today we never use version
9.0 and thus we always early exit in those tests.
All the tables (target, source or any CTE present) in the SQL statement are local i.e. a merge-sql with a combination of Citus local and
Non-Citus tables (regular Postgres tables) should work and give the same result as Postgres MERGE on regular tables. Catch and throw an
exception (not-yet-supported) for all other scenarios during Citus-planning phase.
DESCRIPTION: Support ALTER TABLE .. ADD PRIMARY KEY ... command
Before processing
> **ALTER TABLE ... ADD PRIMARY KEY ...**
command
1. Create a primary key name to use as the constraint name.
2. Change the **ALTER TABLE ... ADD PRIMARY KEY ...** command to into
**ALTER TABLE ... ADD CONSTRAINT \<constraint name> PRIMARY KEY ...**
form.
This is the only form we can specify a name for a primary key. If we run
ALTER TABLE .. ADD PRIMARY KEY, postgres
would create a constraint name internally in its own scheme. But the
problem is that we need to create constraint names
for shards in our own scheme which is \<constraint name>_\<shardid>.
Hence we need to create a name and send it to workers so that the
workers can append the shardid.
4. Run the changed command on the coordinator to make sure we are using
the same constraint name across the board.
5. Send the changed command to workers such that it is executed for the
main table as well as for the shards.
Fixes#6515.
Removes unused job boundary tag `SUBQUERY_MAP_MERGE_JOB`.
Only usage is at `BuildMapMergeJob`, which is only called when the
boundary = `JOIN_MAP_MERGE_JOB`. Hence, it should be safe to remove.
Fixes#6501
Before this commit, we created an additional WaitEventSet for
checking whether the remote socket is closed per connection -
only once at the start of the execution.
However, for certain workloads, such as pgbench select-only
workloads, the creation/deletion of the additional WaitEventSet
adds ~7% CPU overhead, which is also reflected on the benchmark
results.
With this commit, we use the same WaitEventSet for the purposes
of checking the remote socket at the start of the execution.
We use "rebuildWaitEventSet" flag so that the executor can re-use
the existing WaitEventSet.
As a result, we see the following improvements on PG 15:
main : 120051 tps, 0.532 ms latency avg.
avoid_wes_rebuild: 127119 tps, 0.503 ms latency avg.
And, on PG 14, as expected, there is no difference
main : 129191 tps, 0.495 ms latency avg.
avoid_wes_rebuild: 129480 tps, 0.494 ms latency avg.
But, note that PG 15 is slightly (~1.5%) slower than PG 14.
That is probably the overhead of checking the remote socket.
Before this commit, we created an additional WaitEventSet for
checking whether the remote socket is closed per connection -
only once at the start of the execution.
However, for certain workloads, such as pgbench select-only
workloads, the creation/deletion of the additional WaitEventSet
adds ~7% CPU overhead, which is also reflected on the benchmark
results.
With this commit, we use the same WaitEventSet for the purposes
of checking the remote socket at the start of the execution.
We use "rebuildWaitEventSet" flag so that the executor can re-use
the existing WaitEventSet.
As a result, we see the following improvements on PG 15:
main : 120051 tps, 0.532 ms latency avg.
avoid_wes_rebuild: 127119 tps, 0.503 ms latency avg.
And, on PG 14, as expected, there is no difference
main : 129191 tps, 0.495 ms latency avg.
avoid_wes_rebuild: 129480 tps, 0.494 ms latency avg.
But, note that PG 15 is slightly (~1.5%) slower than PG 14.
That is probably the overhead of checking the remote socket.
Fixes a missed include in #6315.
While adding the cluster clock we have added some extra steps to
`citus_prepare_pg_upgrade` and `citus_finish_pg_upgrade`. These changes
were not added to the citus upgrade and downgrade scripts, this allowed
for a syntax error to slip in.
This PR adds the new versions of both UDF's to the upgrade script while
adding the old version to the downgrade script. This exposed the syntax
error which is also solved.
- Because of the make command used for vanilla tests, test status is
always shown as success on CI. As a fix, I added `&& false` at the end
of the copying diff file to make the command fail when check-vanilla
fails.
```make
check-vanilla: all
$(pg_regress_multi_check) --vanillatest || (cp $(vanilla_diffs_file) $(citus_abs_srcdir)/regression.diffs && false)
```
- I also fixed some vanilla tests that fails due to recently added clock
related operators shown up at some queries.
We already have citus_job_wait to wait until the job reaches the desired
state. That PR adds waiting on task state to allow more granular
waiting. It can be used for Citus operations. Moreover, it is also
useful for testing purposes. (wait until a task reaches specified state)
Related to #6459.
Fixes task executor SIGTERM handling.
Problem:
When task executors are sent SIGTERM, their default handler
`bgworker_die`, which is set at worker startup, logs FATAL error. But
they do not release locks there before logging the error, which
sometimes causes hanging of the monitor. e.g. Monitor waits for the lock
forever at pg_stat flush after calling proc_exit.
Solution:
Because executors have connection to backend, they should handle SIGTERM
similar to normal backends. Normal backends uses `die` handler, in which
they set ProcDiePending flag and the next CHECK_FOR_INTERRUPTS call
handles it gracefully by releasing any lock before termination.
This PR adds a new CI workflow named ```flaky-test``` to run flaky test
detection on newly introduced regression tests.
Co-authored-by: Jelte Fennema <github-tech@jeltef.nl>
Adding a testing function `wait_for_resource_cleanup` which waits until
all records in `pg_dist_cleanup` are cleaned up. The motivation is to
prevent flakiness in our tests, since the `NOTICE: cleaned up X orphaned
resources` message is not consistent in many cases. This PR replaces
`citus_cleanup_orphaned_resources` calls with
`wait_for_resource_cleanup` calls.
DESCRIPTION: Adds support for outer joins having a recurring rel in the
outer side of the join (e.g., \<reference table\> LEFT JOIN
\<distributed table\>)
Closes#6219.
Closes#521
If the outer part of an outer join is a recurring rel (i.e., reference
table
or an intermediate_result injected into the query during the earlier
stages
of the recursive planning), Citus cannot run the join query if the other
side
of the join is not a recurring rel (i.e., distributed table).
See DeferredErrorIfUnsupportedRecurringTuplesJoin for the reasoning.
And to support such joins, now we start recursively planning distributed
side
of such joins so that non-recurring rel becomes an intermediate result
(and
hence a recurring rel) since Citus already knows how to compute an outer
join
between two recurring rels already. In the simplest scenario, this means
to
convert
_"\<reference\> LEFT JOIN \<distributed\>"_ to
_"\<reference\> LEFT JOIN \<intermediate_result\>"_
by wrapping the distributed table into a subquery.
- [x] Add support for outer joins having a recurring rel in the outer
side and a "distributed table" (*) in the inner side of the join
- [x] Expand "distributed table" concept to "distributed rel" in first
item.
This means that;
- [x] Currently RecursivelyPlanNonRecurringJoinNode doesn't know how to
wrap a sub join tree that constitutes a recurring rel, e.g., rhs clause
of the following join: `reference LEFT OUTER <distributed INNER JOIN
distributed>`; fix this.
- [x] Similar to previous item, currently
RecursivelyPlanNonRecurringJoinNode doesn't know how to handle
subqueries constituting a distributed rel, e.g., `SELECT * FROM ref LEFT
JOIN (SELECT * FROM dist_1) u1 ON (ref.a = u1.a);`; fix this.
- [x] Add lateral join checks for now-supported outer joins into
recursive planner
- [x] Fix regressions tests
- [x] Verified each test output file by first un-distributing Citus
tables involved in related queries and re-running the test file.
- [x] Some of the tests --that were not supposed to return any data
before but this PR adds support for-- were likely to get flaky, so added
some "ORDER BY"s to them.
- [x] Continue doing manual testing and start writing a test file for
the join clauses that this PR adds support for --not only rely on
existing tests
See https://github.com/citusdata/citus/issues/6546 for what we could do
further.
DESCRIPTION: Create replication artifacts with unique names
We're creating replication objects with generic names. This disallows us
to enable parallel shard moves, as two operations might use the same
objects. With this PR, we'll create below objects with operation
specific names, by appending OparationId to the names.
* Subscriptions
* Publications
* Replication Slots
* Users created for subscriptions
1) Regular users fail to use clock UDF with permission issue.
2) Clock functions were declared as STABLE, whereas by definition they are VOLATILE. By design, any clock/time
functions will return different results for each call even within a single SQL statement.
Note: UDF citus_get_transaction_clock() is a misnomer as it internally calls the clock tick which always returns
different results for every invocation in the same transaction.
Adds signal handlers for graceful termination, cancellation of
task executors and detecting config updates. Related to PR #6459.
#### How to handle termination signal?
Monitor need to gracefully terminate all running task executors before
terminating. Hence, we have sigterm handler for the monitor.
#### How to handle cancellation signal?
Monitor need to gracefully cancel all running task executors before
terminating. Hence, we have sigint handler for the monitor.
#### How to detect configuration changes?
Monitor has SIGHUP handler to reflect configuration changes while
executing tasks.
Emel Şimşek
aykut-bozkurt
Önder Kalacı
Ahmet Gedemenli
Jelte Fennema
Naisila Puka
Hanefi Onaldi
Teja Mupparti
Nils Dijk
Gokhan Gulbiz
Gürkan İndibay
dependabot[bot]
Onur Tirtir