On 8/20/09 1:24 AM, Mike Naberezny wrote:
> Hi Roger,
>
> Roger Hoover wrote:
>> I submitted a patch on Sunday.  I just wanted to check if it got caught
>> by a spam filter or if you guys didn't get a chance to look at it yet.
>
> We're sorry for the delay in responding.  We did receive your patch.
>
>> Since then, I thought of one more idea to improve it and wanted to get
>> your feedback.  The way the patch works now is that a reference count is
>> keep at the FastCGIProcessGroup level and when each fcgi subprocess in
>> spawned and reaped, the increments or decrements the reference count.
>> When the count hits zero, the socket is closed and when it goes from
>> zero to one, it gets created again.
>>
>> The downside to this scheme is that if you restart an fcgi process
>> group, there will be a period of time where the socket is down, which
>> can cause client to to get errors during that time.  The idea I had is
>> that we really only need to tear down the socket when the group config
>> is changed during a reload.  I believe that this always happens though a
>> called to the stopProcessGroup() method in the RPC interface.
>
> That is correct.
>
>> If so, we
>> can have that method trigger a callback on the process group (something
>> like ProcessGroup.stop_requested()).  The base class does not need to do
>> anything in that method but the FastCGIProcessGroup can set a flag that
>> shutdown has been signaled and will close the fcgi socket when that flag
>> is set and the reference count hits zero.
>>
>> If you agree with that concept, that should be a pretty simple change
>> and I can submit a new patch.
>
> This sounds fine to me.  Thanks!

Me too,

- C


>
> Regards,
> Mike
>

_______________________________________________
Supervisor-users mailing list
[email protected]
http://lists.supervisord.org/mailman/listinfo/supervisor-users

Reply via email to