[Gluster-devel] Current state of multi-thread epoll (and review request)

[Shyam]

An update: <inline>

On 01/27/2015 10:14 AM, Shyam wrote:
> Hi,
>
> Here is the current state of multi-thread epoll patch,
>
> 1) Patches:
> epoll patch: http://review.gluster.org/#/c/3842/
> epoll configuration patch: http://review.gluster.org/#/c/9488/
>
> 2) Failures:
> - epoll patch
>    - In 2 runs the following 2 test cases failed in separate instances,
>      - ./tests/bugs/quota/bug-1038598.t
>        - Under investigation

Pending...

>      - ./tests/bugs/glusterd/bug-1104642.t

Fixed, here: http://review.gluster.org/#/c/9498/

>        - This is possibly a TC issue, as we check peer count and then
> immediately check for the configuration update. A peer count does not
> guarantee that the peer has also updated configuration. I will be
> sending a patch (today) to address this test case.
>
> - epoll configuration
>   - Overall this patch passed, could be a one off as it depends on the
> above patch, or could be due to the fact that this makes glusterd and
> the CLI single threaded, and hence we do not hit the above failures.
>
> - NetBSD
>    - NetBSD fails badly with the above patch
>      - Under investigation

Should be fixed, with the latest revision of the MT epoll patch. 
Patchset 16 contains reasons for failure as comments, and patchset 17 
fixes the same.

>
> 3) Requests
> - Request that people review the above patches, so that we can get
> concerns addressed quicker
>
> - The new patch on the configuration of epoll threads, needs some
> attention in terms of correctness of layers where this is allowed to be
> configured (among others) etc. Request reviewers to take a look at this.
>
> Thanks,
> Shyam
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-devel