Re: Out of Memory errors are frustrating as heck!

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

From: Tom Lane <tgl@sss.pgh.pa.us>
To: Jeff Janes <jeff.janes@gmail.com>
Cc: Gunther <raj@gusw.net>, Justin Pryzby <pryzby@telsasoft.com>, Tomas Vondra <tomas.vondra@2ndquadrant.com>, Alvaro Herrera <alvherre@2ndquadrant.com>, pgsql-performance@lists.postgresql.org
Date: 2019-04-22T17:57:34Z
Lists: pgsql-performance
Jeff Janes <jeff.janes@gmail.com> writes:
> Is there a reason to not just elog the HJDEBUG stuff?

Yes --- it'd be expensive (a "no op" elog is far from free) and
useless to ~ 99.999% of users.

Almost all the conditionally-compiled debug support in the PG executor
is legacy leftovers from Berkeley days.  If it were useful more often
than once in a blue moon, we probably would have made it more easily
reachable long ago.  I'm a bit surprised we haven't just ripped it
out, TBH.  When I find myself needing extra debug output, it's almost
never the case that any of that old code does what I need.

There might be a case for changing it all to print to stderr not
stdout, so that it plays more nicely with elog/ereport output
when you do have it turned on, though.

			regards, tom lane



Commits

  1. Consider BufFiles when adjusting hashjoin parameters

  2. Allocate hash join files in a separate memory context