pg_sleep_enhancements.patch

Hello

I am looking on this patch

http://www.postgresql.org/message-id/525FE206.6000502@dalibo.com

a) pg_sleep_for - no objection - it is simple and secure

b) pg_sleep_until

I am not sure - maybe this implementation is too simply. I see two possible
risk where it should not work as users can expect

a) what will be expected behave whem time is changed - CET/CEST ?

b) what will be expected behave when board clock is not accurate and it is
periodically fixed (by NTP) - isn't better to sleep only few seconds and
recalculate sleeping interval?

Regards

Pavel

On 01/29/2014 08:04 PM, Pavel Stehule wrote:
> Hello
>
> I am looking on this patch

Thank you for looking at it.

> http://www.postgresql.org/message-id/525FE206.6000502@dalibo.com
>
> a) pg_sleep_for - no objection - it is simple and secure

Okay.

> b) pg_sleep_until
>
> I am not sure - maybe this implementation is too simply. I see two
> possible risk where it should not work as users can expect
>
> a) what will be expected behave whem time is changed - CET/CEST ?

There is no risk there, the wake up time is specified with time zone.

> b) what will be expected behave when board clock is not accurate and
> it is periodically fixed (by NTP) - isn't better to sleep only few
> seconds and recalculate sleeping interval?

We could do that, but it seems like overkill. It would mean writing a
new C function whereas this is just a simple helper for the existing
pg_sleep() function. So my vote is to keep the patch as-is.

--
Vik

2014-01-29 Vik Fearing <vik(dot)fearing(at)dalibo(dot)com>

> On 01/29/2014 08:04 PM, Pavel Stehule wrote:
> > Hello
> >
> > I am looking on this patch
>
> Thank you for looking at it.
>
> > http://www.postgresql.org/message-id/525FE206.6000502@dalibo.com
> >
> > a) pg_sleep_for - no objection - it is simple and secure
>
> Okay.
>
> > b) pg_sleep_until
> >
> > I am not sure - maybe this implementation is too simply. I see two
> > possible risk where it should not work as users can expect
> >
> > a) what will be expected behave whem time is changed - CET/CEST ?
>
> There is no risk there, the wake up time is specified with time zone.
>
> > b) what will be expected behave when board clock is not accurate and
> > it is periodically fixed (by NTP) - isn't better to sleep only few
> > seconds and recalculate sleeping interval?
>
> We could do that, but it seems like overkill. It would mean writing a
> new C function whereas this is just a simple helper for the existing
> pg_sleep() function. So my vote is to keep the patch as-is.
>
>
Ok

second question - is not this functionality too dangerous? If somebody use
it as scheduler, then

a) can holds connect, session data, locks too long time
b) it can stop on query timeout probably much more early then user expect

What is expected use case?

Regards

Pavel

> --
> Vik
>
>

On 01/29/2014 08:21 PM, Pavel Stehule wrote:
> second question - is not this functionality too dangerous? If somebody
> use it as scheduler, then
>
> a) can holds connect, session data, locks too long time
> b) it can stop on query timeout probably much more early then user expect
>
> What is expected use case?

It is no more dangerous than plain pg_sleep(). The use case is
convenience and clarity of code.

I don't think people will be using it as a scheduler any more than they
do with pg_sleep() because it can't cross transaction boundaries.

--
Vik

2014-01-29 Vik Fearing <vik(dot)fearing(at)dalibo(dot)com>

> On 01/29/2014 08:21 PM, Pavel Stehule wrote:
> > second question - is not this functionality too dangerous? If somebody
> > use it as scheduler, then
> >
> > a) can holds connect, session data, locks too long time
> > b) it can stop on query timeout probably much more early then user expect
> >
> > What is expected use case?
>
> It is no more dangerous than plain pg_sleep(). The use case is
> convenience and clarity of code.
>
> I don't think people will be using it as a scheduler any more than they
> do with pg_sleep() because it can't cross transaction boundaries.
>

I am sure so experienced user didn't use it. But beginners can be messed
due similarity with schedulers.

I invite a) documented these risks b) opinion of other hackers

Probably when it is used as single statement in transaction, then it can
breaks only vacuum

Pavel

>
> --
> Vik
>
>