Re: BUG #13970: Vacuum hangs on particular table; cannot be terminated - requires `kill -QUIT pid`

Brian Ghidinelli wrote:

> Alvaro, thanks for the quick response. I’ve done some additional research and found your involvement on other
threadsin 2014 and 2015 that have allowed me to collect more supporting details. In a nutshell, it appears to be
multixidrelated and /maybe/ pg_upgrade related; I don’t know if we used it on the 9.3 branch or not. I don’t have
thedebug symbols but I did attach gdb and could see it was working on multixid. I don’t have the ability to compile
fromsource on this production server unfortunately but I’m happy to do whatever else I can to help the team find a
solution.

Okay, great.  What would be most helpful is figuring out the pg_upgrade
history of this server; if you have copies of the cluster just before
the upgrade, to extract the "nextMultiXactId" value, that would be
useful.

How large is this table?  We could try to scan it to look for the values
that are causing the problem, and set oldestMxact to one that would not
cause a problem.

How large is the cluster?  For experimentation, it would be very useful
if you could take a copy of it, on a server where you could recompile
with debugging symbols.

--
Álvaro Herrera                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services