browser.cache.disk.capacity is set early in new profiles, causing quick session split and environment-change ping in telemetry-tests-client tests
Categories
(Toolkit :: Telemetry, defect, P1)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox68 | --- | fixed |
People
(Reporter: chutten, Assigned: chutten)
References
Details
Attachments
(3 files)
+++ This bug was initially created as a clone of Bug #1545207 +++
browser.cache.disk.capacity is a userPref in that it is included in the Telemetry Environment and is monitored for changes so as to produce a session split and a "main" ping with reason "environment-change". In the event the smart cache sizer dictates a disk cache size different from the default, this is changed.
In normal execution this happens close enough to the start of a session that the minimum subsession length squelches the ping before it is even assembled.
In telemetry-tests-client the minimum subsession length is set to 0 so there is no squelch.
My Windows laptop, for whatever reason, gets a different-from-default cache size set, causing the pref change causing the extra ping causing the tests to fail.
This bug is about figuring out what to do about this.
|
Assignee | |
Comment 1•7 years ago
|
||
Looks like we can turn the smart disk cache sizer off with browser.cache.disk.smart_size.enabled. Let's do that.
|
|
Assignee | |
Comment 2•7 years ago
|
||
This is because the smart sizer changes a monitored userPref close to
startup, which induces an extra "main" ping (causing tests to fail).
|
||
Comment 4•7 years ago
|
||
Backed out changeset 9b16812e723e (Bug 1547750) for telemetry failes at test_subsession_management.py.
Backout: https://hg.mozilla.org/integration/autoland/rev/574edb245bd0df86ad2705b78461fb72ae0f2428
Push that started the failures: https://treeherder.mozilla.org/#/jobs?repo=autoland&revision=9b16812e723e3dff41c118cd9bcd1f7535452f36&selectedJob=243413066
Failure log: https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=243413066&repo=autoland&lineNumber=1816
|
|
Assignee | |
Comment 5•7 years ago
|
||
There seems to be a timing failure causing install_addon to not always cause an "environment-change"-reason "main" ping. Near as I can tell from the Telemetry tracelogs, this is because the Telemetry Environment hasn't finished initializing so squelches the early changes.
This might help explain why the failures happened on debug and asan builds: slower builds are more likely to not have finished init.
I was able to reproduce bug 1547750 locally on my Windows laptop with a debug build.
We need a plan, I think.
|
|
Assignee | |
Updated•7 years ago
|
|
|
Assignee | |
Comment 6•7 years ago
|
||
This sometimes introduces an environment change ping when it shouldn't, and
hides that install_addon doesn't actually trigger an environment change.
Depends on D29388
|
|
Assignee | |
Comment 7•7 years ago
|
||
|
|
Assignee | |
Comment 8•7 years ago
|
||
Alrighty, I think we've nailed it. Turns out the disk cache capacity change was triggering the environment change. When I removed it, it showed that install_addon didn't actually cause the needed ping reliably.
So now we ensure TelemetryEnvironment's been init'd before installing the addon.
Try run: https://treeherder.mozilla.org/#/jobs?repo=try&revision=d0a36165af121feb2b738911e5e21c097322fd12
(ignore the flake8 failure, that's in a WIP optout ping test patch for bug 1539830)
|
|
||
Comment 10•7 years ago
|
||
| bugherder | ||
https://hg.mozilla.org/mozilla-central/rev/9491a79c5266
https://hg.mozilla.org/mozilla-central/rev/b97e4935d9b9
Description
•