Hi,
First, thank you for your insightful discussion.
I remade patches and attached in this mail.
> From: Tsunakawa, Takayuki
> OTOH, it may be better to commit the tcp_user_timeout patch when
> Nagaura-san has refined the documentation, and then continue
> socket_timeout.
Yes, I want to commit TCP_USER_TIMEOUT patches in PG12.
Also I'd like to continue to discuss about socket_timeout after this CF.
# Thank you for your answering.
About TCP_USER_TIMEOUT:
1) Documentation and checking on UNIX system
As far as I surveyed (solaris and BSD family), these UNIX OS don't have "TCP_USER_TIMEOUT" parameter.
Accordingly I have not checked on real machine.
Also, I modified documentations to remove "equivalent to socket option"
As for socket_timeout:
1) documentation
> From: Robert Haas <robertmhaas@gmail.com>
> This can't be used to force a global query timeout, because any kind of protocol
> message - e.g. a NOTICE or WARNING - would reset the timeout, allowing the
> continue past the supposedly-global query timeout. There may be some other
> ways that can happen, too, either currently or in the future.
Understood. Indeed, doc was not correct.
Mikalai-san,
You used to advise me about via UNIX domain sockets, but it turned out to be able to work.
Therefore, the documentation doesn't have the sentence about it.
Mikalai-san and Tsunakawa-san,
I modified documentation based on your comments about socket_timeout > other parameters.
2) checking whether int type
I Implemented in the current patch. Please see line 73 and 85 in the patch.
3) setting inheritance by "\c" command.
Inheritance of optional parameters such as this parameter,
(but not basic connection parameters such as dbname,)
should be implemented as one feature of "\c".
It is because, seen from users,
it is strange that some parameters can be taken over while the others can't.
Sorry for my late reply again and again...
Best regards,
---------------------
Ryohei Nagaura