Horton

Re: Decoding speculative insert with toast leaks memory

Dilip Kumar <dilipbalaut@gmail.com>

View thread View on postgr.es
From: Dilip Kumar <dilipbalaut@gmail.com>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Andres Freund <andres@anarazel.de>, Petr Jelinek <petr.jelinek@enterprisedb.com>, Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Date: 2021-05-27T04:05:37Z
Lists: pgsql-hackers 
On Thu, May 27, 2021 at 9:03 AM Amit Kapila <amit.kapila16@gmail.com> wrote:
>
> On Thu, Mar 25, 2021 at 11:04 AM Ashutosh Bapat
> <ashutosh.bapat.oss@gmail.com> wrote:
> >
> > Hi All,
> > We saw OOM in a system where WAL sender consumed Gigabttes of memory
> > which was never released. Upon investigation, we found out that there
> > were many ReorderBufferToastHash memory contexts linked to
> > ReorderBuffer context, together consuming gigs of memory. They were
> > running INSERT ... ON CONFLICT .. among other things. A similar report
> > at [1]
> >
> ..
> >
> > but by then we might have reused the toast_hash and thus can not be
> > destroyed. But that isn't the problem since the reused toast_hash will
> > be destroyed eventually.
> >
> > It's only when the change next to speculative insert is something
> > other than INSERT/UPDATE/DELETE that we have to worry about a
> > speculative insert that was never confirmed. So may be for those
> > cases, we check whether specinsert != null and destroy toast_hash if
> > it exists.
> >
>
> Can we consider the possibility to destroy the toast_hash in
> ReorderBufferCleanupTXN/ReorderBufferTruncateTXN? It will delay the
> clean up of memory till the end of stream or txn but there won't be
> any memory leak.

Currently, we are ignoring XLH_DELETE_IS_SUPER, so maybe we can do
something based on this flag?


-- 
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com

Commits

  1. Fix ABI break introduced by commit 4daa140a2f.

  2. Fix valgrind issue in pgoutput.c.

  3. Fix decoding of speculative aborts.

  4. pgoutput: Fix memory leak due to RelationSyncEntry.map.

  5. Fix initialization of RelationSyncEntry for streaming transactions.

  6. Implement streaming mode in ReorderBuffer.