From: | Nikolay Samokhvalov <samokhvalov(at)gmail(dot)com> |
---|---|
To: | Andrey Borodin <amborodin86(at)gmail(dot)com> |
Cc: | Nathan Bossart <nathandbossart(at)gmail(dot)com>, Andres Freund <andres(at)anarazel(dot)de>, Michael Paquier <michael(dot)paquier(at)gmail(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: Transaction timeout |
Date: | 2023-01-13 16:03:00 |
Message-ID: | CANNMO+K01WNSddTOVF6hAo1VGA4K0mFd7RFBBjzSUYNqCGRt_g@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On Thu, Jan 12, 2023 at 11:47 AM Andrey Borodin <amborodin86(at)gmail(dot)com>
wrote:
> On Thu, Jan 12, 2023 at 11:24 AM Nathan Bossart
> <nathandbossart(at)gmail(dot)com> wrote:
> >
> > On Sun, Dec 18, 2022 at 12:53:31PM -0800, Andrey Borodin wrote:
> > > I've rewritten this part to correctly report all timeouts that did
> > > happen. However there's now a tricky comma-formatting code which was
> > > tested only manually.
>
Testing it again, a couple of questions
1) The current test set has only 2 simple cases – I'd suggest adding one
more (that one that didn't work in v1):
gitpod=# set transaction_timeout to '20ms';
SET
gitpod=# begin; select pg_sleep(.01); select pg_sleep(.01); select
pg_sleep(.01); commit;
BEGIN
pg_sleep
----------
(1 row)
ERROR: canceling statement due to transaction timeout
gitpod=# set statement_timeout to '20ms'; set transaction_timeout to 0; --
to test value for statement_timeout and see that it doesn't fail
SET
SET
gitpod=# begin; select pg_sleep(.01); select pg_sleep(.01); select
pg_sleep(.01); commit;
BEGIN
pg_sleep
----------
(1 row)
pg_sleep
----------
(1 row)
pg_sleep
----------
(1 row)
COMMIT
2) Testing for a longer transaction (2 min), in a gitpod VM (everything is
local, no network involved)
// not sure what's happening here, maybe some overheads that are not
related to the implementation,
// but the goal was to see how precise the limiting is for longer
transactions
gitpod=# set transaction_timeout to '2min';
SET
gitpod=# begin;
BEGIN
gitpod=*# select now(), clock_timestamp(), pg_sleep(3) \watch 1
Fri 13 Jan 2023 03:49:24 PM UTC (every 1s)
now | clock_timestamp | pg_sleep
-------------------------------+-------------------------------+----------
2023-01-13 15:49:22.906924+00 | 2023-01-13 15:49:24.088728+00 |
(1 row)
[...]
Fri 13 Jan 2023 03:51:18 PM UTC (every 1s)
now | clock_timestamp | pg_sleep
-------------------------------+-------------------------------+----------
2023-01-13 15:49:22.906924+00 | 2023-01-13 15:51:18.179579+00 |
(1 row)
ERROR: canceling statement due to transaction timeout
gitpod=!#
gitpod=!# rollback;
ROLLBACK
gitpod=# select timestamptz '2023-01-13 15:51:18.179579+00' - '2023-01-13
15:49:22.906924+00';
?column?
-----------------
00:01:55.272655
(1 row)
gitpod=# select interval '2min' - '00:01:55.272655';
?column?
-----------------
00:00:04.727345
(1 row)
gitpod=# select interval '2min' - '00:01:55.272655' - '4s';
?column?
-----------------
00:00:00.727345
(1 row)
– it seems we could (should) have one more successful "1s wait, 3s sleep"
iteration here, ~727ms somehow wasted in a loop, quite a lot.
From | Date | Subject | |
---|---|---|---|
Next Message | Jelte Fennema | 2023-01-13 17:10:23 | Re: [EXTERNAL] Re: Support load balancing in libpq |
Previous Message | Ted Yu | 2023-01-13 15:37:29 | properly sizing attnums in pg_get_publication_tables |