Horton

Re: LogwrtResult contended spinlock

Alvaro Herrera <alvherre@2ndquadrant.com>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Andres Freund <andres@anarazel.de>
Cc: pgsql-hackers@lists.postgresql.org, Jaime Casanova <jaime.casanova@2ndquadrant.com>
Date: 2021-11-18T22:15:58Z
Lists: pgsql-hackers

Attachments

Here's a further attempt at this.  Sorry it took so long.
In this version, I replaced the coupled-in-a-struct representation of
Write&Flush with two separate global variables.  The reason to do this
is to cater to Andres' idea to keep them up-to-date separately.  Of
course, I could kept them together, but it seems more sensible this way.

Andres also suggested to not use monotonic-advance in XLogWrite, because
these update are always done with WALWriteLock held in exclusive mode.
However, that doesn't work (and tests fail pretty quickly) because
XLogWrite seems to be called possibly out of order, so you still have to
check if the current value is newer than what you have to write anyway;
using stock atomic write doesn't work.

So this is v4.

-- 
Álvaro Herrera           39°49'30"S 73°17'W  —  https://www.EnterpriseDB.com/

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