[users at bb.net] Another anecdote, with possible bug.

[Neil Gilmore]

Hi Pierre,

Thanks for the confirmation. Please note, because I failed to mention 
it, that we set collapseRequests globally for most schedulers.

Our fix ended up being to append the master name to the schedulers. We 
lose failover for the schedulers, though.

I may look at your solution if I get some time. I may not understand 
correctly, but it sounds like I'd end up with a builder with the same 
name on 2 different masters. One would be the actual builder, and the 
other would exist solely for its collapseRequests behavior?

It would seem that an eventual fix would involve using the collapse 
behavior attached specifically to a builder, if present. If not present, 
the global config would be able to use the default function, given 
function, or nothing. I don't think it would be too hard to implement 
that. The only oddity right now is that scheduling has to jump through 
the builder to find the function.

Neil Gilmore
grammatech.com

On 11/23/2016 5:55 AM, Pierre Tardy wrote:
> Hi Neil,
> Thanks for the info, this is indeed a design problem that was unknown 
> to me.
> The collapse function will only work in case of a symmetric 
> multimaster setup.
>
> in 0.8, the collapsing was done at the build start time, this is why 
> the config was done in the builder.
> Now it is made at the buildrequest insertion time so that the 
> collapsing algorithm can be more efficient
>
> You can workaround this by configuring dummy builderconfig with only 
> your collapse algorithm (and perhaps attach a dummy slave that is 
> meant to be always disconnected)
>
> Pierre
>
> Le mar. 22 nov. 2016 à 23:20, Neil Gilmore <ngilmore at grammatech.com 
> <mailto:ngilmore at grammatech.com>> a écrit :
>
>     Hi everyone,
>
>     We might have a possible multi-master bug. Maybe it's already known.
>
>     It appears that if collapseRequests is active, buildrequests will only
>     collapse if the scheduler and builder are both on the same master.
>     Otherwise, it appears as though the collapse function will never be
>     found, and the build will not collapse.
>
>     We're getting some requests queueing up that we didn't want.
>
>     Neil Gilmore
>     grammatech.com <http://grammatech.com>
>     _______________________________________________
>     users mailing list
>     users at buildbot.net <mailto:users at buildbot.net>
>     https://lists.buildbot.net/mailman/listinfo/users
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.buildbot.net/pipermail/users/attachments/20161128/9febc51f/attachment.html>