[vlendec at samba.org: [SCM] Samba Shared Repository - branch master updated]

[Jeremy Allison]

On Tue, Mar 01, 2011 at 11:34:51AM +1100, tridge at samba.org wrote:
> Hi Metze,
> 
>  > I hope to get some time in the next days to think about a more generic
>  > fallback from epoll to poll in the standard backend.
> 
> We could also drop the fallback, and just have the epoll(), poll() and
> select() backends as separate backends, defaulting at to epoll() first
> if available, then poll() then select().
> 
> epoll() is no longer an unusual call, at least on Linux. If the system
> has epoll() then expecting it to work is not unreasonable these
> days. If there are other systems that have epoll() but it returns
> -ENOSYS, then we could detect that at startup and auto-switch to the
> next backend.
> 
> I also think that some of the issues we had with epoll() in the early
> days were probably related to the unusual fork() behaviour. I think
> we've got that all sorted out now (with the addition of the pid checks
> in the standard and epoll backends).

Works for me. Removing the fallback gets rid of all the worry about
debugging strange live customer cases :-).

Jeremy.