Re: Perform streaming logical transactions by background workers and parallel apply
Peter Smith <smithpb2250@gmail.com>
Here are my review comments for the patch v23-0003:
======
3.1. src/backend/replication/logical/applybgworker.c -
apply_bgworker_relation_check
+ * Although the commit order is maintained by only allowing one process to
+ * commit at a time, the access order to the relation has changed. This could
+ * cause unexpected problems if the unique column on the replicated table is
+ * inconsistent with the publisher-side or contains non-immutable functions
+ * when applying transactions using an apply background worker.
+ */
+void
+apply_bgworker_relation_check(LogicalRepRelMapEntry *rel)
I’m not sure, but should that second sentence be rearranged as follows?
SUGGESTION
This could cause unexpected problems when applying transactions using
an apply background worker if the unique column on the replicated
table is inconsistent with the publisher-side, or if the relation
contains non-immutable functions.
~~~
3.2.
+ if (!am_apply_bgworker() &&
+ (list_length(ApplyBgworkersFreeList) == list_length(ApplyBgworkersList)))
+ return;
Previously I posted I was struggling to understand the above
condition, and then it was explained (see [1] comment #4) that:
> We need to check this for apply bgworker. (Both lists are "NIL" in apply bgworker.)
I think that information should be included in the code comment.
======
3.3. src/include/replication/logicalrelation.h
+/*
+ * States to determine if changes on one relation can be applied using an
+ * apply background worker.
+ */
+typedef enum ParallelApplySafety
+{
+ PARALLEL_APPLY_UNKNOWN = 0,
+ PARALLEL_APPLY_SAFE,
+ PARALLEL_APPLY_UNSAFE
+} ParallelApplySafety;
+
3.3a.
The enum value PARALLEL_APPLY_UNKNOWN doesn't really mean anything.
Maybe naming it PARALLEL_APPLY_SAFETY_UNKNOWN gives it the intended
meaning.
3.3b.
+ PARALLEL_APPLY_UNKNOWN = 0,
I didn't see any reason to explicitly assign this to 0.
~~~
3.4. src/include/replication/logicalrelation.h
@@ -31,6 +42,8 @@ typedef struct LogicalRepRelMapEntry
Relation localrel; /* relcache entry (NULL when closed) */
AttrMap *attrmap; /* map of local attributes to remote ones */
bool updatable; /* Can apply updates/deletes? */
+ ParallelApplySafety parallel_apply; /* Can apply changes in an apply
+
(Similar to above comment #3.3a)
The member name 'parallel_apply' doesn't really mean anything. Perhaps
renaming this to 'parallel_apply_safe' or 'parallel_safe' etc will
give it the intended meaning.
------
[1] https://www.postgresql.org/message-id/OS3PR01MB6275739E73E8BEC5D13FB6739E6B9%40OS3PR01MB6275.jpnprd01.prod.outlook.com
Kind Regards,
Peter Smith.
Fujitsu Australia
Commits
GET /api/v1/messages/:b64id/commits
the thread's linked commits as JSON, with link sources.
API reference →
-
Fix invalid memory access during the shutdown of the parallel apply worker.
- 3d144c6c8602 16.0 landed
-
Fix assertion failure in apply worker.
- de63f8dadee4 16.0 landed
-
Use elog to report unexpected action in handle_streamed_transaction().
- 781ac42d43ab 16.0 landed
-
Use appropriate wait event when sending data in the apply worker.
- d9d7fe68d35e 16.0 landed
-
Allow the logical_replication_mode to be used on the subscriber.
- 9f2213a7c575 16.0 landed
-
Rename GUC logical_decoding_mode to logical_replication_mode.
- 1e8b61735cfb 16.0 landed
-
Display the leader apply worker's PID for parallel apply workers.
- d540a02a724b 16.0 landed
-
Improve the code to decide and process the apply action.
- c981d9145dea 16.0 landed
-
Document the newly added wait events added by commit 216a784829.
- cd06ccd78fcf 16.0 landed
-
Perform apply of large transactions by parallel workers.
- 216a784829c2 16.0 landed
-
Wake up a subscription's replication worker processes after DDL.
- c6e1f62e2cee 16.0 cited
-
Add copyright notices to meson files
- 8284cf5f746f 16.0 cited
-
Better document logical replication parameters
- a8500750ca0a 16.0 cited
-
Add a common function to generate the origin name.
- 776e1c8a5d14 16.0 cited
-
Harmonize parameter names in storage and AM code.
- bfcf1b34805f 16.0 cited
-
Avoid using list_length() to test for empty list.
- efd0c16becbf 16.0 cited
-
Improve two comments related to a boolean DefElem's value
- 8445f5a21d40 16.0 cited
-
Fix partition table's REPLICA IDENTITY checking on the subscriber.
- 26b3455afaa9 15.0 cited
-
Fix data inconsistency between publisher and subscriber.
- b7658c24c7c1 15.0 cited
-
Fix cache look-up failures while applying changes in logical replication.
- 5a97b1325453 15.0 cited