Re: Race conditions in logical decoding

On 2026-Jan-22, Álvaro Herrera wrote:

> On 2026-Jan-22, Antonin Houska wrote:

> > > Do you mean replace
> > > > if (unlikely(!TransactionIdDidCommit(builder->committed.xip[i])))
> > > to
> > > > if (unlikely(TransactionIdIsInProgress(builder->committed.xip[i]) || !TransactionIdDidCommit(builder->committed.xip[i])))
> >
> > This way the synchronous replication gets stuck, as it did when I tried to use
> > XactLockTableWait(): subscriber cannot confirm replication of certain LSN
> > because publisher is not able to even finalize the commit (due to the waiting
> > for the subscriber's confirmation), and therefore publisher it's not able to
> > decode the data and send it to the subscriber.

BTW, the reason XactLockTableWait and TransactionIdIsInProgress() cause
a deadlock in the same way, is that they are using essentially the same
mechanism. The former uses the Lock object on the transaction, which is
released (by the ResourceOwnerRelease(RESOURCE_RELEASE_LOCKS) call in
CommitTransaction) after RecordTransactionCommit() has returned -- that
is, after the wait on a synchronous replica has happened.

XactLockTableWait does an _additional_ test for
TransactionIdIsInProgress, but that should be pretty much innocuous at
that point.

One thing that I just realized I don't know, is what exactly are the two
pieces that are deadlocking. I mean, one is this side that's decoding
commit. But how/why is that other side, the one trying to mark the
transaction as committed, waiting on the commit decoding? Maybe there's
something that we need to do to _that_ side to prevent the blockage, so
that we can use TransactionIdIsInProgress() here.

--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/