On 06/24/2013 09:06 PM, Jason Edgecombe wrote:
On 06/24/2013 02:42 PM, Michael Meffie wrote:
On Mon, 24 Jun 2013 17:00:34 +0100
Simon Wilkinson <[email protected]> wrote:

On 24 Jun 2013, at 16:08, Michael Meffie wrote:
In addition to more cycles per seconds in the pool of build slaves, we
could make the buildbot master configuration a bit smarter about about
scheduling builds from gerrit.
The biggest speed improvement we could make with buildbot would be finding
some way of cancelling any outstanding builds for patchsets 1..n when
patchset n+1 is submitted for a change. The usual way that we end up with day long backlogs is because a large tree of changes is being repeatedly rebased.

I'd think this could be done with a change to the buildbot contributed
GerritScheduler (or maybe even just a "change filter" to the scheduler).

Unfortunately, I've not been able to actually locate the code for the
GerritScheduler, which I believe is currently being used. The information on
wiki.openafs.org seems to be out of date:

    http://wiki.openafs.org/buildbotmasternotes/

Jason, Is this code available somewhere?
The currently-deployed gerrit scheduler code has vanished from the author's github repo. I could make it available, but there isn't much value in doing so. Currently, there are three implementations of the gerrit scheduler:

g1: the currently deployed scheduler, written by Jay Soffian. It was written before gerrit had upstream support. g2: the gerrit scheduler from newer buildbot versions. This one does one gerrit comment per builder...not desirable
g3: a custom version from YFS.

I would like to merge g2 and g3, but I'm waiting for a working config file so that I can move forward. This situation is why I haven't upgraded the buildbot master software.

What ever is done, should be done after getting a workable gerrit scheduler and upgrading.

To clarify, I'm waiting on a co-worker to reconstruct his old buildbot config so that I can compare g2 & g3.
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel

Reply via email to