Horton

Re: LogwrtResult contended spinlock

Alvaro Herrera <alvherre@alvh.no-ip.org>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Jeff Davis <pgsql@j-davis.com>
Cc: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, Andres Freund <andres@anarazel.de>, pgsql-hackers@lists.postgresql.org, Jaime Casanova <jcasanov@systemguards.com.ec>
Date: 2024-04-07T12:19:29Z
Lists: pgsql-hackers 
I pushed the "copy" pointer now, except that I renamed it to "insert",
which is what we call the operation being tracked.  I also added some
comments.

One perhaps significant change from what Bharath submitted is that we
now use the return value from monotonic advance to return an updated
value of finishedUpto from WaitXLogInsertionsToFinish.  I think this is
a useful optimization: the only caller of WaitXLogInsertionsToFinish
which cares about its return value is XLogFlush, which AFAICS benefits
from having a more up-to-date value.

Also, I did adopt your (Jeff's) idea of using a barrier when reading
logInsertResult in WaitXLogInsertionsToFinish.  If I understand
correctly, this is actually useful: if some other CPU core was also
recently busy waiting for xlog insertions and wrote an updated version
of logInsertResult that's past what we last saw, we benefit from an
updated value because we can skip the spinlock dance and WAL-insert-var
stuff altogether.  With no barrier, we would potentially see an outdated
value of logInsertResult and waste a lot of effort to get it updated.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/
"Those who use electric razors are infidels destined to burn in hell while
we drink from rivers of beer, download free vids and mingle with naked
well shaved babes." (http://slashdot.org/comments.pl?sid=44793&cid=4647152)

Commits

  1. Remove bogus assertion in pg_atomic_monotonic_advance_u64

  2. Add XLogCtl->logInsertResult

  3. Operate XLogCtl->log{Write,Flush}Result with atomics

  4. Split XLogCtl->LogwrtResult into separate struct members

  5. Introduce atomic read/write functions with full barrier semantics.

  6. Reduce the number of GetFlushRecPtr() calls done by walsenders.

  7. Remove most volatile qualifiers from xlog.c