On 2015-05-06 20:55:30 +0000, daniel@heroku.com wrote:
> Postgres gets stuck in this loop and, not checking for signals, will not
> respect the SIGTERM from pg_terminate_backend. This seems like a variant of
> other problems I've seen with pg_terminate_backend not being able to
> function when one is stuck in Openssl, though it seems that in recent
> versions of Postgres that PG provides its own low level I/O functions (e.g.
> my_sock_write) and maybe this is fixable?
I hope I've fixed this for 9.5. The changes are too invasive to
backpatch unfortunately.
> #0 0x00007fe42be95462 in send () from /lib/x86_64-linux-gnu/libc.so.6
> #1 0x00007fe42dcbc487 in my_sock_write (h=0x7fe42efa5eb0, buf=<optimized
> out>, size=<optimized out>)
> at
> /tmp/buildd/postgresql-9.4-9.4.1/build/../src/backend/libpq/be-secure.c:516
> #2 0x00007fe42cdd4667 in BIO_write () from
> /lib/x86_64-linux-gnu/libcrypto.so.1.0.0
> #3 0x00007fe42d0f0260 in ?? () from /lib/x86_64-linux-gnu/libssl.so.1.0.0
> #4 0x00007fe42d0f0984 in ?? () from /lib/x86_64-linux-gnu/libssl.so.1.0.0
> #5 0x00007fe42dcbd1d5 in secure_write (port=0x7fe42efa51b0,
> ptr=0x7fe42efaad90, len=<optimized out>)
> at
Yep, we're stuck writing something and we can't, in older versions,
accept cancel interrupts there because we're inside openssl.
Greetings,
Andres Freund