Justin Pryzby <pryzby@telsasoft.com> writes:
> We never had any issue during the ~2 years running PG96 on this VM, until
> upgrading Monday to PG10.1, and we've now hit it 5+ times.
> BTW this is a VM run on a hypervisor managed by our customer:
> DMI: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 06/22/2012
> Linux TS-DB 2.6.32-431.el6.x86_64 #1 SMP Fri Nov 22 03:15:09 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
Actually ... I was focusing on the wrong part of that. It's not
your hypervisor, it's your kernel. Running four-year-old kernels
is seldom a great idea, and in this case, the one you're using
contains the well-reported missed-futex-wakeups bug:
https://bugs.centos.org/view.php?id=8371
While rebuilding PG so it doesn't use POSIX semaphores will dodge
that bug, I think a kernel update would be a far better idea.
There are lots of other known bugs in that version.
Relevant to our discussion, the fix involves inserting a memory
barrier into the kernel's futex call handling:
https://github.com/torvalds/linux/commit/76835b0ebf8a7fe85beb03c75121419a7dec52f0
regards, tom lane