Closed Bug 900545 Opened 11 years ago Closed 11 years ago

PGO-only talos bustage on inbound

Categories

(Release Engineering :: General, defect)

defect
Not set
blocker

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: emorley, Assigned: jmaher)

References

Details

Attachments

(2 files)

(In reply to Ed Morley [:edmorley UTC+1] from comment #0)
> Done:
> 1) Retrigger on other trees to see if this was caused by a mozharness change.

mozilla-central:
https://tbpl.mozilla.org/?jobname=pgo.*talos

= green

> 2) Retrigger PGO on pushes inside the regression range above.

These will take 3-4 hours.
I noticed we don't run pgo talos on mozharness, it is run the old way.  I suspect this has some of the reasons for failure.
Oh ha, good spot! :-)
Adding jhopkins as buildduty to be on the loop.

I will look into this.

Talos PGO on central looks good as per retriggers from comment 1.
Component: Talos → Release Engineering: Automation (General)
Product: Testing → mozilla.org
QA Contact: catlee
Version: Trunk → other
Blocks: 713055
Assignee: nobody → armenzg
I know I'm duplicating a block of code but I did not want to try to optimize it while I'm trying to fix the tree.
Comment on attachment 784452 [details] [diff] [review]
[buildbotcustom] duplicate code to create talos ScriptFactory for PGO builders

Review of attachment 784452 [details] [diff] [review]:
-----------------------------------------------------------------

Landed as r=bustage as aki is on PTO and jhopkins could only agree with the plan more than being able to review it.
http://hg.mozilla.org/build/buildbotcustom/rev/00217a62c5bc

I've asked jyeo to clean the code after we get out of the hole rather than having duplicated code.

As soon as we reconfigure, we will trigger talos jobs and see after 15 minutes what results we get.

Let's hope this was the issue.
Attachment #784452 - Flags: checked-in+
Thanks Armen! :-)
We have fixed all other talos jobs.
We still have to fix xperf pgo; It is still running the old way.

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&jobname=Windows%207%2032-bit%20mozilla-inbound%20pgo%20talos%20xperf

TEST-UNEXPECTED-FAIL : xperf: File '{firefox}\omni.ja' was accessed and we were not expecting it.  DiskReadCount: 46, DiskWriteCount: 0, DiskReadBytes: 3014656, DiskWriteBytes: 0
TEST-UNEXPECTED-FAIL : xperf: File '{firefox}\browser\omni.ja' was accessed and we were not expecting it.  DiskReadCount: 28, DiskWriteCount: 0, DiskReadBytes: 1835008, DiskWriteBytes: 0
Return code: 1
is omni.ja only seen on PGO builds?  I haven't seen this in all my testing and the initial deployment of xperf.
Blocks: 900605
(In reply to Armen Zambrano G. [:armenzg] (Release Enginerring) (EDT/UTC-4) from comment #8)
> We still have to fix xperf pgo; It is still running the old way.
>
I don't know how I thought that it was still running the old code.

Do we want to close this bug and file another one for the xperf pgo issue?
Assignee: armenzg → jmaher
On PGO builds we optimize the omni.ja such that we can readahead part of it when it it first opened. I suspect that, since those reads occur earlier in the browser execution, xperf sees this as a new read during startup. We should whitelist the omni.ja files.
I wonder how TalosFactory was not letting us see this.

It seems that the issue is also on Mozilla-Central.
I've re-triggered all xperf pgo jobs back to last Friday.
https://tbpl.mozilla.org/?jobname=Windows%207%2032-bit%20mozilla-central%20pgo%20talos%20xperf&showall=1
Attachment #784539 - Flags: review?(aklotz) → review+
landed on talos, will deploy tomorrow morning:
http://hg.mozilla.org/build/talos/rev/f3e761110635
Original issue fixed, closing.
Status: NEW → RESOLVED
Closed: 11 years ago
Resolution: --- → FIXED
Blocks: 900913
Blocks: 901715
Product: mozilla.org → Release Engineering
Component: General Automation → General
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: