Horton

Re: pgBufferUsage.blk_{read|write}_time are zero although there are pgBufferUsage.local_blks_{read|written}

Michael Paquier <michael@paquier.xyz>

View thread
From: Michael Paquier <michael@paquier.xyz>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Nazir Bilal Yavuz <byavuz81@gmail.com>, Melanie Plageman <melanieplageman@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2023-10-10T00:54:43Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix description of I/O timing info for shared buffers in EXPLAIN (BUFFERS)

  2. pg_stat_statements: Add local_blk_{read|write}_time

  3. Add local_blk_{read|write}_time I/O timing statistics for local blocks

  4. Rename I/O timing statistics columns to shared_blk_{read|write}_time

  5. Count write times when extending relation files for shared buffers

  6. Add JIT deform_counter

On Thu, Oct 05, 2023 at 08:51:40AM -0400, Robert Haas wrote:
> On Thu, Oct 5, 2023 at 6:25 AM Nazir Bilal Yavuz <byavuz81@gmail.com> wrote:
>> What do you think about the second patch, counting extend calls'
>> timings in blk_write_time? In my opinion if something increments
>> {shared|local}_blks_written, then it needs to be counted in
>> blk_write_time too. I am not sure why it is decided like that.
> 
> I agree that an extend should be counted the same way as a write. But
> I'm suspicious that here too we have confusion about whether
> blk_write_time is supposed to be covering shared buffers and local
> buffers or just shared buffers.

Agreed.

In ~14, as far as I can see blk_write_time is only incremented for
shared buffers.  FWIW, I agree that we should improve these stats for
local buffers but I am not on board with a solution where we'd use the
same counter for local and shared buffers while we've historically
only counted the former, because that could confuse existing
monitoring queries.  It seems to me that the right solution is to do
the same separation as temp blocks with two separate counters, without
a backpatch.  I'd like to go as far as renaming blk_read_time and
blk_write_time to respectively shared_blk_read_time and
shared_blk_write_time to know exactly what the type of block dealt
with is when querying this data, particularly for pg_stat_statements's
sake.
--
Michael