Horton

Re: Condition variable live lock

Robert Haas <robertmhaas@gmail.com>

View thread View on postgr.es
From: Robert Haas <robertmhaas@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Thomas Munro <thomas.munro@enterprisedb.com>, Andres Freund <andres@anarazel.de>, Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2018-01-05T19:52:38Z
Lists: pgsql-hackers 
On Fri, Jan 5, 2018 at 2:38 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> * I think the Asserts in ConditionVariablePrepareToSleep and
> ConditionVariableSleep ought to be replaced by full-fledged test and
> elog(ERROR), so that they are enforced even in non-assert builds.
> I don't have a lot of confidence that corner cases that could violate
> those usage restrictions would get caught during developer testing.
> Nor do I see an argument that we can't afford the cycles to check.

I think those usage restrictions are pretty basic, and I think this
might be used in some places where performance does matter.  So -1
from me for this change.

> * ConditionVariablePrepareToSleep needs to be rearranged so that failure
> to create the WaitEventSet doesn't leave us in an invalid state.

+1.

> * A lot of the comments could be improved, IMHO.

No opinion without seeing what you propose to change.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

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.