Re: dblink doesn't honor interrupts while waiting a result

Hi

dblink in 8.3 blocks without any possibility of interrupting it while
waiting for an answer from the remote server. Here is a strace
[pid 27607] rt_sigaction(SIGPIPE, {SIG_IGN}, {SIG_IGN}, 8) = 0
[pid 27607] sendto(56, "Q\0\0\0008lock table travelhit.booking_code in
exclusive mode\0", 57, 0, NULL, 0) = 57
[pid 27607] rt_sigaction(SIGPIPE, {SIG_IGN}, {SIG_IGN}, 8) = 0
[pid 27607] poll([{fd=56, events=POLLIN|POLLERR}], 1, -1) = ?
ERESTART_RESTARTBLOCK (To be restarted)
[pid 27607] --- SIGTERM (Terminated) @ 0 (0) ---
[pid 27607] rt_sigreturn(0xf) = -1 EINTR (Interrupted system call)
[pid 27607] poll(

As you can see I'm trying to lock the table travelhit.booking_code,
which blocks because someone else is already holding that lock. When
I send a SIGTERM to the backend, the poll() syscalll is interruped -
but immediatly restarted.

I'm not sure how a proper fix for this could look like, since the
blocking actually happens inside libpq - but this certainly makes
working with dblink painfull...

regards, Florian Pflug

On 2/25/08, Florian G. Pflug <fgp(at)phlo(dot)org> wrote:
> I'm not sure how a proper fix for this could look like, since the
> blocking actually happens inside libpq - but this certainly makes
> working with dblink painfull...

Proper fix would be to use async libpq API, then loop on poll(2)
with small timeout. You can look at pl/proxy for example code.

--
marko

Marko Kreen wrote:
> On 2/25/08, Florian G. Pflug <fgp(at)phlo(dot)org> wrote:
>> I'm not sure how a proper fix for this could look like, since the
>> blocking actually happens inside libpq - but this certainly makes
>> working with dblink painfull...
>
> Proper fix would be to use async libpq API, then loop on poll(2)
> with small timeout. You can look at pl/proxy for example code.
Ah, cool, I'll check out pl/proxy.

regards, Florian Pflug

On Mon, Feb 25, 2008 at 04:45:43AM +0100, Florian G. Pflug wrote:
> dblink in 8.3 blocks without any possibility of interrupting it while
> waiting for an answer from the remote server. Here is a strace
> [pid 27607] rt_sigaction(SIGPIPE, {SIG_IGN}, {SIG_IGN}, 8) = 0
> [pid 27607] sendto(56, "Q\0\0\0008lock table travelhit.booking_code in
> exclusive mode\0", 57, 0, NULL, 0) = 57
> [pid 27607] rt_sigaction(SIGPIPE, {SIG_IGN}, {SIG_IGN}, 8) = 0
> [pid 27607] poll([{fd=56, events=POLLIN|POLLERR}], 1, -1) = ?
> ERESTART_RESTARTBLOCK (To be restarted)
> [pid 27607] --- SIGTERM (Terminated) @ 0 (0) ---
> [pid 27607] rt_sigreturn(0xf) = -1 EINTR (Interrupted system call)
> [pid 27607] poll(
>
> As you can see I'm trying to lock the table travelhit.booking_code,
> which blocks because someone else is already holding that lock. When
> I send a SIGTERM to the backend, the poll() syscalll is interruped -
> but immediatly restarted.
>
> I'm not sure how a proper fix for this could look like, since the
> blocking actually happens inside libpq - but this certainly makes
> working with dblink painfull...

FWIW, I've had problems with dblink getting wedged somewhere in it's
communication with another server. I'm not sure at what stage this
happens (getting results, sending a query, etc). The only way I've found
to clear it is to restart the database the connection was coming from.
Dunno if this is related or not...
--
Decibel!, aka Jim C. Nasby, Database Architect decibel(at)decibel(dot)org
Give your computer some brain candy! www.distributed.net Team #1828