Horton

Re: Parallel scan with SubTransGetTopmostTransaction assert coredump

Maxim Orlov <m.orlov@postgrespro.ru>

View thread View on postgr.es
From: Maxim Orlov <m.orlov@postgrespro.ru>
To: Greg Nancarrow <gregn4422@gmail.com>
Cc: Pavel Borisov <pashkin.elfe@gmail.com>, Michael Paquier <michael@paquier.xyz>, Pengchengliu <pengchengliu@tju.edu.cn>, Andres Freund <andres@anarazel.de>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2021-06-04T12:30:15Z
Lists: pgsql-hackers

Attachments

Just a note here. After examining the core dump I did notice something.

While in XidInMVCCSnapshot call the snapshot->suboverflowed is set true 
although subxip == NULL and subxcnt == 0. As far as I understand, 
snapshot->suboverflowed is set true in the GetRunningTransactionData 
call.

And then I decided to put elog around CurrentRunningXacts->subxcnt's 
assigment.
diff --git a/src/backend/storage/ipc/procarray.c 
b/src/backend/storage/ipc/procarray.c
index 42a89fc5dc9..3d2db02f580 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2781,6 +2781,9 @@ GetRunningTransactionData(void)
         * increases if slots do.
         */

+       if (suboverflowed)
+               elog(WARNING, " >>> CurrentRunningXacts->subxid_overflow 
is true");
+
        CurrentRunningXacts->xcnt = count - subcount;
        CurrentRunningXacts->subxcnt = subcount;
        CurrentRunningXacts->subxid_overflow = suboverflowed;

... and did get a bunch of messages. I.e. subxid_overflow is set true 
very often.

I've increased the value of PGPROC_MAX_CACHED_SUBXIDS. Once it becomes 
more than 120 there are no messages and no failed assertions are 
provided any more.

---
Best regards,
Maxim Orlov.

Commits

  1. Fix broken snapshot handling in parallel workers.

  2. Doc: move some catalogs.sgml entries to the right place.

  3. Stamp 13.2.

  4. Create an infrastructure for parallel computation in PostgreSQL.