Horton

Re: TRAP: FailedAssertion("prev_first_lsn < cur_txn->first_lsn", File: "reorderbuffer.c", Line: 927, PID: 568639)

Amit Kapila <amit.kapila16@gmail.com>

View thread View on postgr.es
From: Amit Kapila <amit.kapila16@gmail.com>
To: Masahiko Sawada <sawada.mshk@gmail.com>
Cc: Tomas Vondra <tomas.vondra@enterprisedb.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2022-10-18T10:55:54Z
Lists: pgsql-hackers 
On Mon, Oct 17, 2022 at 7:05 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> On Thu, Oct 13, 2022 at 4:08 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
> >
> > --- a/src/backend/replication/logical/decode.c
> > +++ b/src/backend/replication/logical/decode.c
> > @@ -113,6 +113,15 @@
> > LogicalDecodingProcessRecord(LogicalDecodingContext *ctx,
> > XLogReaderState *recor
> >   buf.origptr);
> >   }
> >
> > +#ifdef USE_ASSERT_CHECKING
> > + /*
> > + * Check the order of transaction LSNs when we reached the start decoding
> > + * LSN. See the comments in AssertTXNLsnOrder() for details.
> > + */
> > + if (SnapBuildGetStartDecodingAt(ctx->snapshot_builder) == buf.origptr)
> > + AssertTXNLsnOrder(ctx->reorder);
> > +#endif
> > +
> >   rmgr = GetRmgr(XLogRecGetRmid(record));
> > >
> >
> > I am not able to think how/when this check will be useful. Because we
> > skipped assert checking only for records that are prior to
> > start_decoding_at point, I think for those records ordering should
> > have been checked before the restart. start_decoding_at point will be
> > either (a) confirmed_flush location, or (b) lsn sent by client, and
> > any record prior to that must have been processed before restart.
>
> Good point. I was considering the case where the client sets far ahead
> LSN but it's not worth considering this case in this context. I've
> updated the patch accoringly.
>

One minor comment:
Can we slightly change the comment: ". The ordering of the records
prior to the LSN, we should have been checked before the restart." to
". The ordering of the records prior to the start_decoding_at LSN
should have been checked before the restart."?

-- 
With Regards,
Amit Kapila.

Commits

  1. Fix executing invalidation messages generated by subtransactions during decoding.

  2. Fix assertion failures while processing NEW_CID record in logical decoding.

  3. Immediately WAL-log subtransaction and top-level XID association.

  4. Stop demanding that top xact must be seen before subxact in decoding.