Обсуждение: How Many PG_Locks are considered too many

On 7/30/15 6:13 AM, Renato Oliveira wrote:
> We have a Nagios plugin, which monitors pg_locks and almost daily we see
> 3000 to 40000 pg_locks.
>
> Can we just ignore them, can we let them grow without worrying?
>
> How many pg_locks are considered unsafe for any given postgres server?

That depends on how many concurrent clients you have and what they are
doing.  Every table access will at least create a share lock of some
kind, so if you have a lot of activity that does a lot of things, you
will see a lot of locks, but that doesn't impact database performance in
a significant way.

I don't think monitoring the absolute number of locks is useful.  You
might want to chart it, to compare over time.  If you want to monitor
locks, you could monitor lock waits, which you can get by checking the
server log.

Peter thank you much appreciated

Sent from my iPhone

> On 30 Jul 2015, at 14:54, Peter Eisentraut <peter_e@gmx.net> wrote:
>
>> On 7/30/15 6:13 AM, Renato Oliveira wrote:
>> We have a Nagios plugin, which monitors pg_locks and almost daily we see
>> 3000 to 40000 pg_locks.
>>
>> Can we just ignore them, can we let them grow without worrying?
>>
>> How many pg_locks are considered unsafe for any given postgres server?
>
> That depends on how many concurrent clients you have and what they are
> doing.  Every table access will at least create a share lock of some
> kind, so if you have a lot of activity that does a lot of things, you
> will see a lot of locks, but that doesn't impact database performance in
> a significant way.
>
> I don't think monitoring the absolute number of locks is useful.  You
> might want to chart it, to compare over time.  If you want to monitor
> locks, you could monitor lock waits, which you can get by checking the
> server log.
>

On Thu, Jul 30, 2015 at 9:19 AM, John Scalia <jayknowsunix@gmail.com> wrote:
> Seconding Peter on this one; it's a lot more important should one of those
> locks be hanging around, say for hours or days, not how many have come and
> gone.

Also, it's good to focus on *ungranted* locks.   Typically the only
time I care about granted locks is to find out which process is
keeping my other process getting its lock granted.

merlin

Thank you appreciated

Sent from my iPhone

> On 30 Jul 2015, at 20:05, Merlin Moncure <mmoncure@gmail.com> wrote:
>
>> On Thu, Jul 30, 2015 at 9:19 AM, John Scalia <jayknowsunix@gmail.com> wrote:
>> Seconding Peter on this one; it's a lot more important should one of those
>> locks be hanging around, say for hours or days, not how many have come and
>> gone.
>
> Also, it's good to focus on *ungranted* locks.   Typically the only
> time I care about granted locks is to find out which process is
> keeping my other process getting its lock granted.
>
> merlin
>