Merlin Moncure <mmoncure@gmail.com> wrote:
> Takahiro<itagaki.takahiro@oss.ntt.co.jp> wrote:
> > contrib/dblink seems to have no treatments for query cancels.
> > (1) Users need to wait for completion of remote query.
> > (2) PGresult objects will be memory leak.
Here is a patch to fix the issues. I hope the fixes will be ported
to older versions if possible.
(1) is fixed by using non-blocking APIs in libpq. I think we should
always use non-blocking APIs even if the dblink function itself is
a blocking-function.
(2) is fixed by RegisterXactCallback(AtEOXact_dblink). However, there
might be any better solutions -- for example, ResourceOwner framework.
> > For (1), asynchronous libpq functions should be used instead of blocking
> > ones, and wait for the remote query using a loop with CHECK_FOR_INTERRUPTS().
>
> How would you structure this loop exactly?
Please check execute_query() and wait_for_result() in the patch.
Regards,
---
ITAGAKI Takahiro
NTT Open Source Software Center