Horton

Re: SegFault on 9.6.14

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@gmail.com>
Cc: Amit Kapila <amit.kapila16@gmail.com>, Jerry Sievers <gsievers19@comcast.net>, Tomas Vondra <tomas.vondra@2ndquadrant.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Robert Haas <robertmhaas@gmail.com>
Date: 2019-07-18T13:45:06Z
Lists: pgsql-hackers 
Thomas Munro <thomas.munro@gmail.com> writes:
> Hmm, so something like a new argument "bool final" added to the
> ExecXXXShutdown() functions, which receives false in this case to tell
> it that there could be a rescan so keep the parallel context around.

I think this is going in the wrong direction.  Nodes should *always*
assume that a rescan is possible until ExecEndNode is called.  See the
commentary about EXEC_FLAG_REWIND in executor.h:

 * REWIND indicates that the plan node should try to efficiently support
 * rescans without parameter changes.  (Nodes must support ExecReScan calls
 * in any case, but if this flag was not given, they are at liberty to do it
 * through complete recalculation.  Note that a parameter change forces a
 * full recalculation in any case.)

If nodeLimit is doing something that's incompatible with that, it's
nodeLimit's fault; and similarly for the parallel machinery.

If you want to do otherwise, you are going to be inventing a whole
bunch of complicated and doubtless-initially-buggy control logic
to pass down information about whether a rescan might be possible.
That doesn't sound like a recipe for a back-patchable fix.  Perhaps
we could consider redesigning the rules around REWIND in a future
version, but that's not where to focus the bug fix effort.

			regards, tom lane

Commits

  1. Don't shut down Gather[Merge] early under Limit.

  2. Prohibit shutting down resources if there is a possibility of back up.

  3. Fix buffer usage stats for parallel nodes.

  4. Fix crashes on plans with multiple Gather (Merge) nodes.

  5. Separate reinitialization of shared parallel-scan state from ExecReScan.