R: CachedPlan logs until full disk

Dear Tom,

thank you for the reply.
Tonight this problem happened again:

>> CachedPlan: 1024 total in 1 blocks; 640 free (0 chunks); 384 used
>>     CachedPlanSource: 1024 total in 1 blocks; 336 free (0 chunks); 688 used
>>     SPI Plan: 1024 total in 1 blocks; 912 free (0 chunks); 112 used
>>     CachedPlan: 1024 total in 1 blocks; 200 free (0 chunks); 824 used
>>     CachedPlanSource: 1024 total in 1 blocks; 96 free (0 chunks); 928 used
>>     SPI Plan: 1024 total in 1 blocks; 928 free (0 chunks); 96 used
>>     CachedPlan: 1024 total in 1 blocks; 640 free (0 chunks); 384 used

>This appears to be a fragment of a memory map that would be produced
>in conjunction with an "out of memory" error.  It's difficult to say
>much more than that with only this much information, but clearly you
>need to do something to prevent recurrent out-of-memory errors.

We were automatically bulkling a table that archive system logging.
We used a pg_bulk, but not a create or replace function.

That process often use lots of memory.

Just one question: do you think is it possible to disable that logging sentence?

Thank you!
Francesco

________________________________________
Da: Tom Lane [tgl@sss.pgh.pa.us]
Inviato: venerdì 4 novembre 2016 21.24
A: Job
Cc: pgsql-general@postgresql.org
Oggetto: Re: [GENERAL] CachedPlan logs until full disk

Job <Job@colliniconsulting.it> writes:
> it is the second time (in two weeks), that  have a very strange Postgresql in a 8.4.22 installation (32 bit still).

You realize, of course, that 8.4.x has been out of support for a couple of
years now.

> Logfile grow up (in few hours) until filling the Whole disk space.
> I can read infinite series of this messages:

> CachedPlan: 1024 total in 1 blocks; 640 free (0 chunks); 384 used
>     CachedPlanSource: 1024 total in 1 blocks; 336 free (0 chunks); 688 used
>     SPI Plan: 1024 total in 1 blocks; 912 free (0 chunks); 112 used
>     CachedPlan: 1024 total in 1 blocks; 200 free (0 chunks); 824 used
>     CachedPlanSource: 1024 total in 1 blocks; 96 free (0 chunks); 928 used
>     SPI Plan: 1024 total in 1 blocks; 928 free (0 chunks); 96 used
>     CachedPlan: 1024 total in 1 blocks; 640 free (0 chunks); 384 used

This appears to be a fragment of a memory map that would be produced
in conjunction with an "out of memory" error.  It's difficult to say
much more than that with only this much information, but clearly you
need to do something to prevent recurrent out-of-memory errors.

If looking at the map as a whole makes it clear that it's zillions
of CachedPlans that are chewing up most of the memory, then I would
guess that they are getting leaked as a result of constantly replacing
plpgsql functions --- does your application do a lot of
CREATE OR REPLACE FUNCTION commands?  I don't think plpgsql coped
with that very well before 9.1.

                        regards, tom lane