On Thu, 17 Feb 2011 17:37:08 -0000, John A Meinel <email address hidden> wrote:
> We talked about increasing the number of file handles, but when wgrant
> tested that on his machine, it caused problems because 'select' doesn't
> like more than 1024 handles, and python has some internal MAX_FD set, etc.
If we're close to the select() limit in production, we should probably
be using a different reactor *anyway*.
This translates to passing -r poll or -r epoll to twistd, so I think it
would be something the losas need to change in an init script. And test
on staging first, etc etc.
On Thu, 17 Feb 2011 17:37:08 -0000, John A Meinel <email address hidden> wrote:
> We talked about increasing the number of file handles, but when wgrant
> tested that on his machine, it caused problems because 'select' doesn't
> like more than 1024 handles, and python has some internal MAX_FD set, etc.
If we're close to the select() limit in production, we should probably
be using a different reactor *anyway*.
This translates to passing -r poll or -r epoll to twistd, so I think it
would be something the losas need to change in an init script. And test
on staging first, etc etc.
Cheers,
mwh