Horton

Thread

Commits

  1. Avoid "invalid memory alloc request size" while reading pg_stat_activity.

  1. invalid memory alloc request size from pg_stat_activity?

    James Tomson <james@pushd.com> — 2019-05-06T23:04:18Z

    Hello,

We're upgrading from postgresql 9.6 to 10.6 and everything is working fine, but any query against pg_stat_activity results in ERROR:  invalid memory alloc request size 1652113408

e.g:

ourdb=> SELECT pid, state, age(query_start, clock_timestamp()), usename, query FROM pg_stat_activity WHERE query != '<IDLE>' AND state != 'idle' ORDER BY age limit 100;
ERROR:  invalid memory alloc request size 1652113408

Maybe this has to do with us setting track_activity_query_size=102400? Is there a known safe maximum for that, or could there be some other problem?

Thanks,
James

  2. Re: invalid memory alloc request size from pg_stat_activity?

    Euler Taveira <euler@timbira.com.br> — 2019-05-07T13:37:41Z

    Em seg, 6 de mai de 2019 às 20:04, James Tomson <james@pushd.com> escreveu:
>
> ourdb=> SELECT pid, state, age(query_start, clock_timestamp()), usename, query FROM pg_stat_activity WHERE query != '<IDLE>' AND state != 'idle' ORDER BY age limit 100;
> ERROR:  invalid memory alloc request size 1652113408
>
That is because it is limited to 1GB - 1.

> Maybe this has to do with us setting track_activity_query_size=102400? Is there a known safe maximum for that, or could there be some other problem?
>
Question is: why do you want to return such a big query text? That's
impractical for admin/monitoring tools. We can usually identify a
query with 1024 bytes (that is the default).


-- 
   Euler Taveira                                   Timbira -
http://www.timbira.com.br/
   PostgreSQL: Consultoria, Desenvolvimento, Suporte 24x7 e Treinamento

  3. Re: invalid memory alloc request size from pg_stat_activity?

    Alvaro Herrera <alvherre@2ndquadrant.com> — 2019-05-07T14:55:13Z

    On 2019-May-07, Euler Taveira wrote:

> Em seg, 6 de mai de 2019 às 20:04, James Tomson <james@pushd.com> escreveu:
> >
> > ourdb=> SELECT pid, state, age(query_start, clock_timestamp()), usename, query FROM pg_stat_activity WHERE query != '<IDLE>' AND state != 'idle' ORDER BY age limit 100;
> > ERROR:  invalid memory alloc request size 1652113408
> >
> That is because it is limited to 1GB - 1.
> 
> > Maybe this has to do with us setting track_activity_query_size=102400? Is there a known safe maximum for that, or could there be some other problem?
> >
> Question is: why do you want to return such a big query text? That's
> impractical for admin/monitoring tools. We can usually identify a
> query with 1024 bytes (that is the default).

Hmm, but 102400 is only 100kB, nowhere near the 1GB-1 limit, so there's
something odd going on there.

I would start investigating by attaching a debugger and setting a
breakpoint on 'errfinish', then running the above query; past the
backtrace here, as it might be illuminating.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

  4. Re: invalid memory alloc request size from pg_stat_activity?

    Tom Lane <tgl@sss.pgh.pa.us> — 2019-05-07T14:59:30Z

    James Tomson <james@pushd.com> writes:
> We're upgrading from postgresql 9.6 to 10.6 and everything is working fine, but any query against pg_stat_activity results in ERROR:  invalid memory alloc request size 1652113408

> e.g:

> ourdb=> SELECT pid, state, age(query_start, clock_timestamp()), usename, query FROM pg_stat_activity WHERE query != '<IDLE>' AND state != 'idle' ORDER BY age limit 100;
> ERROR:  invalid memory alloc request size 1652113408

That seems odd ...

> Maybe this has to do with us setting track_activity_query_size=102400?

It wouldn't surprise me if reading pg_stat_activity causes allocation of
track_activity_query_size * max_connections bytes, but unless you also
have max_connections set to some pretty insane value, that wouldn't
be enough to fit this observation.

Did you have these parameters set differently in 9.6?

			regards, tom lane

  5. Re: invalid memory alloc request size from pg_stat_activity?

    Tom Lane <tgl@sss.pgh.pa.us> — 2019-05-07T15:14:27Z

    Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> Hmm, but 102400 is only 100kB, nowhere near the 1GB-1 limit, so there's
> something odd going on there.

I can reproduce the described behavior by also setting max_connections
to something around 16K.

Now, it seems pretty silly to me to be burning in excess of 1GB of shmem
just for the current-query strings, and then that much again in every
backend that reads pg_stat_activity.  But should we be telling people they
can't do it?  I'm working on a patch to use MemoryContextAllocHuge for
the "localactivity" buffer in pgstat_read_current_status.  It might seem
dumb now, but perhaps in ten years it'll be common.

			regards, tom lane

  6. Re: invalid memory alloc request size from pg_stat_activity?

    James Tomson <james@pushd.com> — 2019-05-07T16:22:39Z

    Thanks all - we did have our max_connections set very very high (100k) as we upgraded this db in a load test environment, and did not want to reboot the db after hitting the lower, saner, production limits during load testing.

I am surprised this allocation was taking place for unused connections - I’ve also verified that lowering max_connections to 10K on this instance fixed the issue for us. Much appreciated! 

Best,
James

> On May 7, 2019, at 11:14 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> 
> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
>> Hmm, but 102400 is only 100kB, nowhere near the 1GB-1 limit, so there's
>> something odd going on there.
> 
> I can reproduce the described behavior by also setting max_connections
> to something around 16K.
> 
> Now, it seems pretty silly to me to be burning in excess of 1GB of shmem
> just for the current-query strings, and then that much again in every
> backend that reads pg_stat_activity.  But should we be telling people they
> can't do it?  I'm working on a patch to use MemoryContextAllocHuge for
> the "localactivity" buffer in pgstat_read_current_status.  It might seem
> dumb now, but perhaps in ten years it'll be common.
> 
> 			regards, tom lane