Horton

Re: Condition variable live lock

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Thomas Munro <thomas.munro@enterprisedb.com>
Cc: Andres Freund <andres@anarazel.de>, Robert Haas <robertmhaas@gmail.com>, Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2018-01-05T17:33:13Z
Lists: pgsql-hackers

Attachments

I wrote:
> It's a question of how complicated you're willing to make this
> logic, and whether you trust that we'll be able to test all the
> code paths.

Attached is a patch incorporating all the ideas mentioned in this thread,
except that I think in HEAD we should change both ConditionVariableSignal
and ConditionVariableBroadcast to return void rather than a possibly
misleading wakeup count.  This could be back-patched as is, though.

As I feared, the existing regression tests are not really adequate for
this: gcov testing shows that the sentinel-inserting code path is
never entered, meaning ConditionVariableBroadcast never sees more
than one waiter.  What's more, it's now also apparent that no outside
caller of ConditionVariableSignal ever actually awakens anything.
So I think it'd be a good idea to expand the regression tests if we
can do so cheaply.  Anybody have ideas about that?  Perhaps a new
module under src/test/modules would be the best way?  Alternatively,
we could drop some of the optimization ideas.

BTW, at least on gaur, this does nothing for the runtime of the join
test, meaning I'd still like to see some effort put into reducing that.

			regards, tom lane

Commits

  1. Add WL_EXIT_ON_PM_DEATH pseudo-event.

  2. While waiting for a condition variable, detect postmaster death.

  3. Fix race condition during replication origin drop.

  4. Allow ConditionVariable[PrepareTo]Sleep to auto-switch between CVs.

  5. Cosmetic improvements in condition_variable.[hc].

  6. Improve error detection capability in proclists.

  7. Remove return values of ConditionVariableSignal/Broadcast.

  8. Reorder steps in ConditionVariablePrepareToSleep for more safety.

  9. Rewrite ConditionVariableBroadcast() to avoid live-lock.

  10. Add parallel-aware hash joins.