Обсуждение: \watch 0 or \watch 0.00001 doesn't do what I want

Daniel's post [1] on \watch reminded me of this little issue I bumped into:

I wanted to run a query in a tight loop, without any delay. I tried 
"\watch 0", but it didn't do what I wanted:

postgres=# \watch 0
Wed 09 Oct 2024 16:34:19 EEST (every 1s)

  ?column?
----------
         1
(1 row)

Wed 09 Oct 2024 16:34:20 EEST (every 1s)

  ?column?
----------
         1
(1 row)

^C

Then I tried setting the delay really small, but that didn't do what I 
wanted either:

postgres=# \watch 0.00001
Wed 09 Oct 2024 16:36:45 EEST (every 1e-05s)

  ?column?
----------
         1
(1 row)

^C

It runs the query just once and then hangs forever, until I hit CTRL-C 
to cancel.

[1] 
https://www.postgresql.org/message-id/B2FD26B4-8F64-4552-A603-5CC3DF1C7103%40yesql.se

-- 
Heikki Linnakangas
Neon (https://neon.tech)

On 09/10/2024 16:38, Heikki Linnakangas wrote:
> Daniel's post [1] on \watch reminded me of this little issue I bumped into:
> 
> I wanted to run a query in a tight loop, without any delay. I tried 
> "\watch 0", but it didn't do what I wanted:
> 
> postgres=# \watch 0

Correction: This changed in version 16. It works the way I expected on 
v16, but not in earlier versions.

> Then I tried setting the delay really small, but that didn't do what I wanted either:
> 
> postgres=# \watch 0.00001
> Wed 09 Oct 2024 16:36:45 EEST (every 1e-05s)
> 
>  ?column?
> ----------
>         1
> (1 row)
> 
> ^C
> 
> It runs the query just once and then hangs forever, until I hit CTRL-C to cancel.

This issue is present on newer versions still.

-- 
Heikki Linnakangas
Neon (https://neon.tech)

Heikki Linnakangas <hlinnaka@iki.fi> writes:
> This issue is present on newer versions still.

Here's the problem:

    long        sleep_ms = (long) (sleep * 1000);

If "sleep" is less than 0.0005, sleep_ms rounds to zero, which
results in the subsequent setitimer disarming rather than
arming the interrupt.

There is an uncommented

        if (sleep == 0)
            continue;

in the loop, which I bet some cowboy added to fix the zero-wait
problem you complained of.  But it's doing the wrong thing because
it checks sleep not sleep_ms.

We should change this to test sleep_ms, and we should probably
fix the code that says what the wait interval is to print
sleep_ms/1000.0 not sleep.  And some more comments would be good.

            regards, tom lane