Re: Interruptible sleeps (was Re: CommitFest 2009-07: Yay, Kevin! Thanks, reviewers!)

On Fri, Sep 3, 2010 at 12:10 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>> On Fri, Sep 3, 2010 at 10:07 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> [ shrug... ]  I stated before that the Hot Standby patch is doing
>>> utterly unsafe things in signal handlers.  Simon rejected that.
>>> I am waiting for irrefutable evidence to emerge from the field
>>> (and am very confident that it will be forthcoming...) before
>>> I argue with him further.  Meanwhile, I'm not going to accept anything
>>> unsafe in a core facility like this patch is going to be.
>
>> Oh.  I thought you had ignored his objections and fixed it.  Why are
>> we releasing 9.0 with this problem again?  Surely this is nuts.
>
> My original review of hot standby found about half a dozen things
> I thought were broken:
> http://archives.postgresql.org/pgsql-hackers/2010-05/msg00178.php
> After a *very* long-drawn-out fight I fixed one of them
> (max_standby_delay), largely still over Simon's objections.  I don't
> have the energy to repeat that another half-dozen times, so I'm going
> to wait for the suspected problems to be proven by field experience.

Bummer. Allow me to cast a vote for doing something about the fact
that handle_standby_sig_alarm() thinks it can safely acquire an LWLock
in a signal handler. I think we should be making our decisions on
what to change in the code based on what is technically sound, rather
than based on how much the author complains about changing it. Of
course there may be cases where there is a legitimate difference of
opinion concerning the best way forward, but I don't believe this is
one of them.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company