Re: database vacuum from cron hanging

Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> writes:
> Kevin Grittner wrote:
>> I'm not sure what you mean regarding pg_config -- could you clarify?

> The output of pg_config --configure

Actually I wanted the whole thing, not just --configure (I'm
particularly interested in the CFLAGS setting).

>> Your email came through as I was trying to figure out where to find
>> the core dump. We restarted the server with cassert, and I find this
>> in the log prior to my attempt to vacuum:

> It should be in $PGDATA/core (maybe with some other name depending on
> settings)

If my theory about a bogus increment code sequence is correct, then the
core dump will not tell us anything very interesting anyway --- the trap
will happen when the slower of the two processes tries to remove its
pin, but that's way after the bug happened.

I'm thinking that the easiest way to confirm or disprove this theory is
to examine the assembly code. Please do this:
1. cd into src/backend/storage/buffer directory of build tree.
2. rm bufmgr.o; make
3. Note gcc command issued by make to rebuild bufmgr.o. Cut and
paste, changing -c to -S and removing "-o bufmgr.o" if present.
Keep all the other switches the same.
4. This should produce a file bufmgr.s. Gzip and send to me
(off-list please, it's likely to be large and boring)
Please also confirm exactly which version of bufmgr.c you are working
with --- the $PostgreSQL line near the head of the file will do.

regards, tom lane