Re: RFC: replace pg_stat_activity.waiting with something more descriptive

On Mon, Sep 14, 2015 at 2:25 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
wrote:

> On Mon, Sep 14, 2015 at 3:02 PM, Alexander Korotkov <aekorotkov(at)gmail(dot)com>
> wrote:
>
>> On Sat, Sep 12, 2015 at 3:24 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
>> wrote:
>>
>>> On Fri, Aug 14, 2015 at 7:23 PM, Alexander Korotkov <
>>> aekorotkov(at)gmail(dot)com> wrote:
>>> >
>>> > On Thu, Aug 6, 2015 at 1:01 PM, Ildus Kurbangaliev <
>>> i(dot)kurbangaliev(at)postgrespro(dot)ru> wrote:
>>> >>
>>> >>
>>> >> I've looked deeper and I found PgBackendStatus to be not a suitable
>>> >> place for keeping information about low level waits. Really,
>>> PgBackendStatus
>>> >> is used to track high level information about backend. This is why
>>> auxiliary
>>> >> processes don't have PgBackendStatus, because they don't have such
>>> information
>>> >> to expose. But when we come to the low level wait events then
>>> auxiliary
>>> >> processes are as useful for monitoring as backends are. WAL writer,
>>> >> checkpointer, bgwriter etc are using LWLocks as well. This is
>>> certainly unclear
>>> >> why they can't be monitored.
>>> >>
>>> >> This is why I think we shoudn't place wait event into
>>> PgBackendStatus. It
>>> >> could be placed into PGPROC or even separate data structure with
>>> different
>>> >> concurrency model which would be most suitable for monitoring.
>>> >
>>> >
>>> > +1 for tracking wait events not only for backends
>>> >
>>> > Ildus, could you do following?
>>> > 1) Extract LWLocks refactoring into separate patch.
>>> > 2) Make a patch with storing current wait event information in PGPROC.
>>> >
>>>
>>> Now as Robert has committed standardization of lwlock names in
>>> commit - aa65de04, let us try to summarize and work on remaining parts
>>> of the patch. So I think now the next set of work is as follows:
>>>
>>> 1. Modify the tranche mechanism so that information about LWLocks
>>> can be tracked easily. For this already there is some discussion, ideas
>>> and initial patch is floated in this thread and there doesn't seem to be
>>> much
>>> conflicts, so we can write the patch for it. I am planning to write or
>>> modify
>>> the existing patch unless you, IIdus or anyone has objections or want to
>>> write it, please let me know to avoid duplication of work.
>>>
>>> 2. Track wait_event in pg_stat_activity. Now here the main point where
>>> we doesn't seem to be in complete agreement is that shall we keep it
>>> as one byte in PgBackendStatus or use two or more bytes to store
>>> wait_event information and still there is another point made by you to
>>> store it in PGProc rather than in PgBackendStatus so that we can display
>>> information of background processes as well.
>>>
>>> Now as a matter of consensus, I think Tom has raised a fair point [1]
>>> against
>>> storing this information in PGProc and I feel that it is relatively less
>>> important to have such information about background processes and the
>>> reason for same is explained upthread [2]. About having more than
>>> one-byte
>>> to store information about various wait_events, I think now we will not
>>> have
>>> more than 100 or so events to track, do you really think that anytime in
>>> forseeable
>>> future we will have more than 256 important events which we would like
>>> to track?
>>>
>>> So I think about this lets first try to build the consensus and then
>>> attempt to
>>> write or modify the patch.
>>>
>>>
>>> [1] - http://www.postgresql.org/message-id/4067.1439561494@sss.pgh.pa.us
>>> [2] -
>>> http://www.postgresql.org/message-id/CAA4eK1JRQGTgRMRao6H65rA=fhifUCNQhX7ONgY58UM6rN1rxQ@mail.gmail.com
>>>
>>
>> In order to build the consensus we need the roadmap for waits monitoring.
>> Would single byte in PgBackendStatus be the only way for tracking wait
>> events? Could we have pluggable infrastructure in waits monitoring: for
>> instance, hooks for wait event begin and end?
>> Limit of 256 wait events is probably OK. But we have no room for any
>> additional parameters of wait event. For instance, if you notice high
>> contention for buffer content lock then it would be nice to figure out: how
>> many blocks are involved?, which blocks? We need to track additional
>> parameters in order to answer this question.
>>
>
> We can track additional parameters by default or based on some
> parameter, but do you think that tracking backends wait_event
> information as proposed hinders in any which way the future extensions
> in this area? The point is that detailed discussion of other parameters
> could be better done separately unless you think that this can block
> future enhancements for waits monitoring.
>

Yes, I think this can block future enhancements for waits monitoring.
You're currently proposing to store current wait event in just single byte.
And this is single byte not because of space economy, it is so by
concurrency design.
Isn't it natural to ask you how are we going to store something more about
current wait event? Should we store additional parameters separately from
current wait event? Does it make sense?
Or should we move current wait event away from BgBackendStatus. If so, why
place it there?

I see wait monitoring of overall
> database as a really valuable feature, but not able to see compelling
> need to sort out everything before this patch. Having said that, I am open
> for discussion if you want specific things to be sorted out before moving
> further.
>

I think we need to be sure that we can move further in waits monitoring
without reverting this.

------
Alexander Korotkov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company