On Wed, May 17, 2017 at 7:24 AM, Kyotaro HORIGUCHI
<horiguchi.kyot...@lab.ntt.co.jp> wrote:
> Hello,
> At Tue, 16 May 2017 12:45:39 -0400, Tom Lane <t...@sss.pgh.pa.us> wrote in 
> <22556.1494953...@sss.pgh.pa.us>
>> Robert Haas <robertmh...@gmail.com> writes:
>> > Concretely, I think we should replace the abort_cleanup_incomplete
>> > flag from my previous patch with a changing_xact_state flag and set
>> > that flag around all transaction state changes, clearing it when such
>> > changes have succeeded.  On error, the flag remains set, so we know
>> > that the state of that connection is unknown and that we must close it
>> > (killing outer transaction levels as needed).
>> > Thoughts?
>> Sounds plausible.
> I think that the current issue is the usability of the current
> connection. Even if any other command should fail, we can
> continue using the connection if ABORT TRANSACTION
> succeeds. Failure of ABORT immediately means that the connection
> is no longer available. If this discuttion is reasonable,
> changing_xact_state might be too-much.

pgfdw_get_result() raises an ERROR, which will end up in
pgfdw_xact_callback with ABORT. There if we found connection in bad
state, we will slam it down. The problem seems to be that waits for
none of the transaction related commands are interruptible; both while
sending the command and while receiving its result. Robert's
suggestion is to fix the later for transaction command. So, it looks
good to me. Can you please elaborate what's too much with
> By the way if an fdw connection is stalled amid do_sql_command
> waiting a result, a cancel request doesn't work (in other words
> pgsql_xact_callback is not called) since EINTR is just ignored in
> libpq. Maybe we should teach libpq to error out with EINTR with
> some additional reasons. PQsetEINTRcallback() or something?

Right do_sql_command() is uninterruptible and so is PQexec() used for
suggestion seems to be that we make them interruptible with
changing_xact_state, so that we could notice a previously failed in
transaction related command in pgfdw_xact_callback(). So, his approach
scores on that front as well.

Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company

Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to