On Tue, Jan 21, 2014 at 8:01 PM, Peter Geoghegan <pg@heroku.com> wrote:
>> BTW shouldn't there be an fflush() in qtext_store?
>
> I don't think so, no. Whenever qtext_store() callers don't fflush()
> before releasing their exclusive lock, they FreeFile() before doing so
> instead. In the case of pgss_shmem_startup() there is no lock held
> anyway, because it doesn't matter, for the same reason it doesn't
> matter that we're modifying shmem structures in a way that ordinarily
> necessitates an exclusive lock. Why fflush() every qtext_store() call
> if there is expected to be more than one, as is the case for several
> callers?
Having said all that, it occurs to me now that I could have been
smarter about where I fflush(). I'm pretty sure pgss_store() should
have two fflush() calls. The first can occur with just a shared LWLock
held, before the lock strength promotion interim (you probably noticed
that I no longer generate a normalized query text in that interim, for
reasons that are obvious). The second fflush() may occur only in the
very unlikely event of a garbage collection necessitating a new
qtext_store() call with an exclusive lock held, a few lines after the
first fflush(). No need to fflush() with the exclusive lock the vast
majority of the time.
--
Peter Geoghegan