Re: long running transactions

[Tom Lane - Tue at 01:18:27PM -0400]
> >> Hmph. Is the autovac process actually doing anything (strace would be
> >> revealing)?

It's definitively doing something; mostly reading, but also some few
writes, semops and opens.

> If not, can you attach to the autovac process with gdb and
> >> get a stack trace to see where it's blocked?

(gdb) bt
#0 0xb7c599f8 in select () from /lib/tls/libc.so.6
#1 0x08253c53 in pg_usleep ()
#2 0x0812ee93 in vacuum_delay_point ()
#3 0x0812f2a5 in lazy_vacuum_rel ()
#4 0x0812ef7b in lazy_vacuum_rel ()
#5 0x0812b4b6 in vac_update_relstats ()
#6 0x0812a995 in vacuum ()
#7 0x0818d2ca in autovac_stopped ()
#8 0x0818ceae in autovac_stopped ()
#9 0x0818c848 in autovac_stopped ()
#10 0x0818c4e2 in autovac_start ()
#11 0x08192c11 in PostmasterMain ()
#12 0x08191dcf in PostmasterMain ()
#13 0x081541b1 in main ()

> It'll come back after the autovacuum naptime. If it gets stuck again,
> please investigate.

It seems stuck, has had the same transid for a long while, and the
number of undeletable dead rows in our tables are increasing.