On Tue, Aug 30, 2016 at 9:34 AM, Maksim Milyutin <m.milyu...@postgrespro.ru>

> On Mon, Aug 29, 2016 at 5:22 PM, maksim <m.milyu...@postgrespro.ru
>> <mailto:m.milyu...@postgrespro.ru>> wrote:
>>     Hi, hackers!
>>     Now I complete extension that provides facility to see the current
>>     state of query execution working on external session in form of
>>     EXPLAIN ANALYZE output. This extension works on 9.5 version, for 9.6
>>     and later it doesn't support detailed statistics for parallel nodes
>> yet.
>>     I want to present patches to the latest version of PostgreSQL core
>>     to enable this extension.
>> Hello,
>> Did you publish the extension itself yet?
> Hello, extension for version 9.5 is available in repository
> https://github.com/postgrespro/pg_query_state/tree/master.
> Last year (actually, exactly one year ago) I was trying to do something
>> very similar, and it quickly turned out that signals are not the best
>> way to make this sort of inspection.  You can find the discussion
>> here: https://www.postgresql.org/message-id/CACACo5Sz7G0MFauC082iM
>> =xx_hq7qq5ndr4jpo+h-o5vp6i...@mail.gmail.com
> Thanks for link!
> My patch *custom_signal.patch* resolves the problem of «heavy» signal
> handlers. In essence, I follow the course offered in *procsignal.c* file.
> They define *ProcSignalReason* values on which the SUGUSR1 is multiplexed.
> Signal recent causes setting flags for *ProcessInterrupt* actuating, i.e.
> procsignal_sigusr1_handler() only sets specific flags. When
> CHECK_FOR_INTERRUPTS appears later on query execution *ProcessInterrupt* is
> called. Then triggered user defined signal handler is executed.
> As a result, we have a deferred signal handling.

Yes, but the problem is that nothing gives you the guarantee that at the
moment you decide to handle the interrupt, the QueryDesc structures you're
looking at are in a good shape for Explain* functions to run on them.  Even
if that appears to be the case in your testing now, there's no way to tell
if that will be the case at any future point in time.

Another problem is use if shm_mq facility, because it manipulates the state
of process latch.  This is not supposed to happen to a backend happily
performing its tasks, at random due to external factors, and this is what
the proposed approach introduces.


Reply via email to