On Sat, Dec 23, 2017 at 09:34:27AM +0530, Amit Kapila wrote:
> On Fri, Dec 22, 2017 at 11:16 PM, Robertson, Alan L
> <Alan.Robertson@charter.com> wrote:
> > Is this described in the documentation somewhere?
>
> Yes. See the documentation of the parameter tcp_keepalives_idle [1].
> It clearly documents the behavior, "In sessions connected via a
> Unix-domain socket, this parameter is ignored and always reads as
> zero".
>
> https://www.postgresql.org/docs/devel/static/runtime-config-connection.html
So the behavior is documented, which is better than leaving it
undocumented, kinda.
The following hack works for me. Apart from being ugly, it's
fundamentally unsatisfactory because the question of whether the
setting has been activated is left unsettled:
SELECT (regexp_match(
pg_read_file(
'postgresql.auto.conf'),
$$tcp_keepalives_idle = '([[:digit:]]+)$$
)
)[1] AS "tcp_keepalives_idle";
tcp_keepalives_idle
---------------------
60
(1 row)
It might be possible to infer the state of activation by riffling
through the logs with similar hackery, but it's tricky to ensure that
the answer you get is current.
Is it worthwhile to reconsider the decision not to expose the
currently active setting of this parameter via UNIX sockets?
Best,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778
Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate
