correct. Also, if the client is still "alive" they have no effect either.
On Fri, May 29, 2009 at 11:01 AM, Y W <luckyasser@gmail.com> wrote:
> I'm sorry that was my bad, they're only ignored when uing Unix-domain
> sockets to connect instead of TCP/IP.
>
> On Fri, May 29, 2009 at 12:55 PM, Y W <luckyasser@gmail.com> wrote:
>>
>> But according to the documentation, are they ignored if postgres was
>> hosted on a Unix/linux system?
>>
>> And are you're pretty sure that connections are terminated on the server
>> side, which will result in releasing any locked resources during that failed
>> transaction?
>>
>> On Fri, May 29, 2009 at 12:22 PM, Scott Marlowe <scott.marlowe@gmail.com>
>> wrote:
>>>
>>> On Fri, May 29, 2009 at 10:18 AM, Y W <luckyasser@gmail.com> wrote:
>>> > Are you referring to connection settings ? Coz if u do, the closet
>>> > thing I
>>> > can find for the tcp_keepalive_timeout u're mentioning are the
>>> > tcp_keepalives_idle, tcp_keepalives_interval, and the
>>> > tcp_keepalives_count.
>>> > And apart from the fact that they're ignored for unix-socket
>>> > connetions,
>>> > these are for terminating the client connection from the client side
>>> > and not
>>> > from the server side.
>>>
>>> Yes those are what I'm referring to, and they apply to the server side
>>> as well. They only terminate DEAD connections. I.e. the client dies,
>>> and x seconds later the connection is closed on the server and the
>>> backend terminated.
>>>
>>> > Otherwise I don't know what are you refering to. I
>>> > want to be able to terminate idle connections from the server side, if
>>> > there
>>> > was a tcp failure with that client which resulted in an unfinished
>>> > transaction and a locked resource, how can u do that ?
>>>
>>> tcp keepalive does that.
>>
>
>
--
When fascism comes to America, it will be intolerance sold as diversity.