488459 - Create 10 new win32 slaves for TryServer pool

Status: RESOLVED FIXED

(Release Engineering :: General, defect, P3)

Description

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

Once the VMs are created, setup the buildbot slaves, and connect them to TryServer.

Comment 1

[[e-mail] (:bhearsum)]

Windows hasn't queued as much on the try server since we added the most recent 4. The past couple days have been elevated, but before that we had nearly a week of no more than 30minutes before build start. production-master is in a similar state - with maybe a slightly higher average. Both have seen elevated usage the past 2 days because of the freeze.

Given that, I think we should look at splitting this 4/6 or 5/5 between the two pools. What do you think, John?

Comment 2

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

(In reply to comment #1)
> Windows hasn't queued as much on the try server since we added the most recent
> 4. The past couple days have been elevated, but before that we had nearly a
> week of no more than 30minutes before build start. production-master is in a
> similar state - with maybe a slightly higher average. Both have seen elevated
> usage the past 2 days because of the freeze.
> 
> Given that, I think we should look at splitting this 4/6 or 5/5 between the two
> pools. What do you think, John?

Its hard to predict. I'd suggest still putting all 10 slaves on TryServer to start with, for the following reasons:
1) right now, both production-master and tryserver have approx same backlog (~5 jobs waited over 15mins in the last 24hours).
2) the performance of the win32 slaves for production-master will improve as their disks gets increased (see bug#489940 for details). This will not be true for win32 slaves on tryserver, so more slaves on TryServer seems correct.
3) during crunch periods, we see a increase in jobs to both production-master and TryServer. However, it seems to me we see a bigger spike on TryServer.

For those reasons, I think it best to put all 10 new slaves on TryServer and watching for a while. They are easy to switch back/forth, and if needed, we can easily rename some later and move them permanently.

Seem reasonable?

Comment 3

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

(In reply to comment #1)
> Windows hasn't queued as much on the try server since we added the most recent
> 4. The past couple days have been elevated, but before that we had nearly a
> week of no more than 30minutes before build start. production-master is in a
> similar state - with maybe a slightly higher average. Both have seen elevated
> usage the past 2 days because of the freeze.
> 
> Given that, I think we should look at splitting this 4/6 or 5/5 between the two
> pools. What do you think, John?

Actually, given the load we saw in April for production-master and TryServer, I'd still like to push these 10 new slaves over to TryServer; the volume of changes was high, none of the changes can be bunched together in TryServer, and there are much fewer slaves on TryServer. Does that seem reasonable?

Comment 4

[[e-mail] (:bhearsum)]

(In reply to comment #3)

> Actually, given the load we saw in April for production-master and TryServer,
> I'd still like to push these 10 new slaves over to TryServer; the volume of
> changes was high, none of the changes can be bunched together in TryServer, and
> there are much fewer slaves on TryServer. Does that seem reasonable?

wfm

Comment 5

[Lukas Blakk [:lsblakk] use ?needinfo]

try-win32-slave10 through 19 are up and running on try-master.