macOS non-qr / non-webrender perf jobs are no longer available, making it harder to assess whether regressions were fixed / how qr compares with non-qr
Categories
(Testing :: Talos, defect, P1)
Tracking
(firefox-esr78 unaffected, firefox82 unaffected, firefox83 unaffected, firefox84 wontfix, firefox85 wontfix, firefox86 wontfix, firefox87 wontfix, firefox88 wontfix, firefox89 fixed)
Tracking | Status | |
---|---|---|
firefox-esr78 | --- | unaffected |
firefox82 | --- | unaffected |
firefox83 | --- | unaffected |
firefox84 | --- | wontfix |
firefox85 | --- | wontfix |
firefox86 | --- | wontfix |
firefox87 | --- | wontfix |
firefox88 | --- | wontfix |
firefox89 | --- | fixed |
People
(Reporter: Gijs, Assigned: davehunt)
References
(Regression)
Details
(Keywords: regression)
Attachments
(1 file, 2 obsolete files)
(deleted),
text/x-phabricator-request
|
Details |
STR:
- update m-c to current tip
./mach try fuzzy --full
- filter for e.g.
'mac 'awsy
ER:
see both qr and non-qr jobs
AR:
no non-qr jobs
Comment 1•4 years ago
|
||
In Fx83 WebRender will be enabled for 90%+ of the macOS population, so for performance purposes we should mostly just care about the QR setup. In the interests of saving resources we switched mac CI to mostly running with WebRender enabled.
Comment 2•4 years ago
|
||
Set release status flags based on info from the regressing bug 1673071
Assignee | ||
Comment 3•4 years ago
|
||
(In reply to Kartikaya Gupta (email:kats@mozilla.staktrace.com) from comment #1)
In Fx83 WebRender will be enabled for 90%+ of the macOS population, so for performance purposes we should mostly just care about the QR setup. In the interests of saving resources we switched mac CI to mostly running with WebRender enabled.
It makes sense that WebRender should be our primary configuration, but should we maintain a subset of tests running with WebRender disabled? A performance regression for ~10% of our users on macOS would still affect a large number of users.
Comment 5•4 years ago
|
||
If we can afford to have some of both that's great. To make a decision here we really need to know how much running both will cost. Dave can you get an estimate for that?
Assignee | ||
Comment 6•4 years ago
|
||
(In reply to Jeff Muizelaar [:jrmuizel] from comment #5)
If we can afford to have some of both that's great. To make a decision here we really need to know how much running both will cost. Dave can you get an estimate for that?
The macOS hardware is more limited, but I suspect we could at least run AWSY. We also have the ability to run some tests less frequently, which could be an option here. I'm not able to answer the question of cost, however I suspect it's more an issue of device pool capacity. Mihai, can you help here?
Comment 7•4 years ago
|
||
Agreed with Joel to 302 this to him. thank you!
Comment 8•4 years ago
|
||
we typically don't have a queue for osx, and we do have a pool that will free up in another week. In short we have capacity to schedule at least 10 jobs, and probably 20 by next week. These will run periodically on autoland and be sheriffed I assume?
Updated•4 years ago
|
Updated•4 years ago
|
Assignee | ||
Comment 9•4 years ago
|
||
(In reply to Joel Maher ( :jmaher ) (UTC -0800) from comment #8)
we typically don't have a queue for osx, and we do have a pool that will free up in another week. In short we have capacity to schedule at least 10 jobs, and probably 20 by next week. These will run periodically on autoland and be sheriffed I assume?
I would suggest AWSY + high value page load tests running on the autoland backstop.
Comment 10•4 years ago
|
||
Let me know if you want me to look into doing this, or if somebody else is going to.
Comment 11•4 years ago
|
||
Is this a wontfix?
Assignee | ||
Comment 12•4 years ago
|
||
(In reply to Ryan VanderMeulen [:RyanVM] from comment #11)
Is this a wontfix?
No, assigning to Kartikaya based on comment 10.
Assignee | ||
Updated•4 years ago
|
Assignee | ||
Comment 13•4 years ago
|
||
:sparky :bebe looks like kats won't be able to work on this, could one of you take a look?
Comment 14•4 years ago
|
||
Updated•4 years ago
|
Comment 15•4 years ago
|
||
(In reply to Dave Hunt [:davehunt] [he/him] ⌚GMT from comment #9)
I would suggest AWSY + high value page load tests running on the autoland backstop.
I wrote a patch that adds back AWSY and raptor-tp6 for non-QR macOS. But I'm not totally sure what you mean by the "autoland backstop". It looks like raptor-tp6 only runs on m-c by default, and awsy runs on autoland/m-c/m-b/m-r. I can trim that down a bit if you want.
Comment 16•4 years ago
|
||
(But also, if the patch is going to be significantly more complicated than what I wrote, it would be better if somebody else picks it up)
Assignee | ||
Comment 17•4 years ago
|
||
(In reply to Kartikaya Gupta (email:kats@mozilla.staktrace.com) from comment #15)
(In reply to Dave Hunt [:davehunt] [he/him] ⌚GMT from comment #9)
I would suggest AWSY + high value page load tests running on the autoland backstop.
I wrote a patch that adds back AWSY and raptor-tp6 for non-QR macOS. But I'm not totally sure what you mean by the "autoland backstop". It looks like raptor-tp6 only runs on m-c by default, and awsy runs on autoland/m-c/m-b/m-r. I can trim that down a bit if you want.
Thanks kats, I appreciate you staying involved! I suspect it might be necessary to get :sparky to help here as he's recently migrated macOS to browsertime. Also, it looks like we might need to do some more work to target the high value tests more easily.
Updated•4 years ago
|
Updated•4 years ago
|
Comment 18•4 years ago
|
||
I'm thinking we should implement a by-tier
key and use it here in the run-on-projects settings for these kinds of cases: https://searchfox.org/mozilla-central/source/taskcluster/ci/test/browsertime-desktop.yml#98-106
Alternatively, we could create a new test group, but then we have the issue that we have two nearly duplicated sets of definitions for these tests which is something we've been trying to avoid.
Another option is to use a transform to modify the run-on-projects settings on tier-1 tests.
:davehunt, I think it might be a good idea for :kimberlythegeek to finish this off since it will get more complicated.
Assignee | ||
Comment 19•4 years ago
|
||
(In reply to Greg Mierzwinski [:sparky] from comment #18)
I'm thinking we should implement a
by-tier
key and use it here in the run-on-projects settings for these kinds of cases: https://searchfox.org/mozilla-central/source/taskcluster/ci/test/browsertime-desktop.yml#98-106Alternatively, we could create a new test group, but then we have the issue that we have two nearly duplicated sets of definitions for these tests which is something we've been trying to avoid.
Another option is to use a transform to modify the run-on-projects settings on tier-1 tests.
I like the idea of a by-tier
if that will work given that the tiers themselves are set by app/subtest.
:davehunt, I think it might be a good idea for :kimberlythegeek to finish this off since it will get more complicated.
:kimberlythegeek could you sync with :sparky on moving this forward?
Comment 20•4 years ago
|
||
Apologies that this slipped through before everyone went on PTO, I will touch base with him as soon as possible.
Updated•4 years ago
|
Comment 21•4 years ago
|
||
Updated•4 years ago
|
Updated•4 years ago
|
Updated•4 years ago
|
Updated•4 years ago
|
Updated•4 years ago
|
Assignee | ||
Comment 22•4 years ago
|
||
Updated•4 years ago
|
Updated•4 years ago
|
Updated•4 years ago
|
Assignee | ||
Updated•4 years ago
|
Updated•4 years ago
|
Updated•4 years ago
|
Comment 23•4 years ago
|
||
Comment 24•4 years ago
|
||
bugherder |
Updated•4 years ago
|
Description
•