Here is a quick patch to allow the socket time out to be configured on a
per-service basis.

Disclaimer: we aren't (yet) running this in production.

--Justin

[ Post a follow-up to this message ]

On 27/06/07, Justin Huff <jjhuff@mspin.net> wrote:
> Here is a quick patch to allow the socket time out to be configured on a
> per-service basis.

Yay, a patch!

> Disclaimer: we aren't (yet) running this in production.

That's probably best, it confuses the two max_idle_time subroutines
that used to be in the codebase ;-)

I've just got rid of one in ClientHTTPBase.pm which is now the
persist_client_timeout option in subversion (that defaults to 30
seconds) and is a timeout in seconds for keep-alives to the end user.

The other one is in Socket.pm: the default is for sockets to never
time out (0). Is this what you are interested in, or does the new
persist_client_timeout configuration option good enough for you?

Regards, Leon

[ Post a follow-up to this message ]

>> Disclaimer: we aren't (yet) running this in production.
> That's probably best, it confuses the two max_idle_time subroutines
> that used to be in the codebase ;-)
Oops! I think I did end up pushing that anyway

> I've just got rid of one in ClientHTTPBase.pm which is now the
> persist_client_timeout option in subversion (that defaults to 30
> seconds) and is a timeout in seconds for keep-alives to the end user.
>
> The other one is in Socket.pm: the default is for sockets to never
> time out (0). Is this what you are interested in, or does the new
> persist_client_timeout configuration option good enough for you?
We aren't using persistent connections since there were some other
issues we ran into.  However, we do have cases where we launch a single
request, and it might sit idle for a while -- perlbal was dropping the
connection.

--Justin

[ Post a follow-up to this message ]