356522 - Turn all tests on argo off

Status: RESOLVED FIXED

(Webtools Graveyard :: Tinderbox, defect)

Description

[Adam Guthrie]

Today around 10:00 PDT Ts went up on argo and Tp saw a slight decrease. No other tinderboxes showed this, and bonsai tells me that no patches went in that could have affected these numbers, so it must be due to the tinderbox itself.

It would be nice to know what causes stuff like this to happen. It screws up perf. numbers and someone going back a month from now and seeing this is not going to know, "Oh, that's when such and such changed; it's really not "real" perf. regression."

Comment 1

[J. Paul Reed [:preed]]

All perf numbers that should be used for anything useful should be coming from the Testing Tinderboxen, not from the build tinderboxen.

The machine for this is: Linux bl-bldlnx01 Dep argo test perf

I'll post to mozilla.dev.builds, but argo's perf numbers should be turned off. (And actually, the story for Linux builds on trunk should be figured out; it looks like both argo and prometheus are doing builds. Anyone know the story behind that?)

Comment 2

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Can we get argo tests turned off asap?  It's reporting what is, as best I can tell, bogus data.  (See dev.performance)

Comment 3

[Adam Guthrie]

We need to figure out what's causing the random spikes in Ts times on bl-bldlnx01 first, IMO.

I talked to rhelmer on IRC about this a few weeks ago. He says he thought it might be due to some type of cron jobs running. He said he was going to turn them off and see if it made a difference.

Comment 4

[J. Paul Reed [:preed]]

Tests on argo are turned off.

ispiked: if there's a spike on bl-bldlnx01, then is there a bug on it with details (when the spike occurs, etc.)?

Comment 5

[Adam Guthrie]

I filed bug 357068 on the spike I mentioned in comment 3.