On Thu, 9 Apr 2015, Royce Ausburn wrote:
> The issue emerged just after we've bumped the max JVM heap size
> significantly and during a heavy running migration/data transformation
> process with lots of jdbc queries.
>
> The issue appears to be in how processDeadParsedQueries() works - if
> there're a lot of recently GC'd parsed statements then there'll be a lot of
> statements to close which manages to fill the send buffer on our end while
> the corresponding postgres process on the server end is busy filling its
> send buffer with responses from the close statement requests.
>
Certainly the driver shouldn't deadlock in this situation and should only
send a limited number of close requests at once, but if your diagnosis is
correct that GC is triggering finalization of these resources, it seems
your application is partly at fault here. If these resources were closed
when complete instead of relying on GC, they would dissipate at a
reasonably rate and not cause this deadlock. Perhaps you need to add some
more close() calls to your code?
Kris Jurka