Update to latest metrics_index list on main#551
Conversation
github-actions
Bot
requested review from
Dexterp37,
akkomar,
badboy,
chutten,
mikaeld,
relud and
whd
as code owners
chutten
force-pushed
the
fog-update/update-metrics-index-main
branch
from
4f28882 to
730e83c
Compare
|
I think these should be fine but I'd appreciate one last approval from @relud before we hit the big green button. Moving and removing files shouldn't be as fraught as moving and removing deps, but it may be more complicated than moving or removing individual metrics definitions. |
| - netwerk/metrics.yaml | ||
| - netwerk/protocol/http/metrics.yaml | ||
| - toolkit/components/cookiebanners/metrics.yaml | ||
| - toolkit/components/crashes/metrics.yaml |
There was a problem hiding this comment.
I'm not 100% sure what happens with apps that rely on this dependency regarding the table-generation, so hearing from @relud would make me feel a lot more comfortable about how this affects that and what problems we might encounter.
My guess is that it should work, since Desktop is still declaring this metrics.yaml file, just in the application definition instead of in this dependency, and the Android apps that consume gecko should still have these metrics definitions from the metrics.yaml file in libcrash. So it should amount to just moving things around, but I don't know if something about how this is generated for storage is affected by this or not.
There was a problem hiding this comment.
As I understand it, due to push mode the results of historical scans will remain in place from metrics|ping|tag files that are no longer referenced in repositories.yaml, so removing these from gecko should be fine.
But, adding these files to an application that depends on gecko would cause duplicated metric identifiers.
After doing some work to fix duplicate metric checks, I was able to produce the following email message in testing:
New Email
From: telemetry-alerts@mozilla.com
To: crash-reporting-wg@mozilla.org,chutten@mozilla.com,stability@mozilla.org,glean-team@mozilla.com
Subject: Glean: Duplicate metric identifiers detected
Body:
Glean has detected duplicated metric identifiers coming from the product 'firefox-desktop'.
- 'crash.process_type' defined more than once in firefox-desktop, gecko
- 'crash.startup' defined more than once in firefox-desktop, gecko
- 'crash.time' defined more than once in firefox-desktop, gecko
- 'crash.uptime' defined more than once in firefox-desktop, gecko
What to do about this:
1. File a bug to track your investigation. You can just copy this email into the bug Description to get you started.
2. Reply-All to this email to let the list know that you are investigating. Include the bug number so we can help out.
3. Rename the most recently added metric to be more specific. See [1]
4. Make sure a Glean team member reviews any patches. Care needs to be taken that the resolution of this problem is schema-compatible.
If you have any problems, please ask for help on the #glean Slack channel. We'll give you a hand.
What this is:
We have a system called probe-scraper [2] that scrapes the metric information from all Mozilla products using the Glean SDK. All the scraped data is available on the probeinfo service [3]. The scraped definition is used to build things such as the probe-dictionary [4] and other data tools. It detected that one metric that was recently added has an identifier collision with some metric that already existed in the application namespace. So it sent this email out, encouraging you to fix the problem.
What happens if you don't fix this:
The metrics will compete to send their data in pings, making the data unreliable at best.
You can do this!
Your Friendly, Neighborhood Glean Team
[1] - https://mozilla.github.io/glean/book/user/adding-new-metrics.html#naming-things
[2] - https://github.com/mozilla/probe-scraper
[3] - https://probeinfo.telemetry.mozilla.org/
[4] - https://telemetry.mozilla.org/probe-dictionary/
There was a problem hiding this comment.
@relud Does it matter that this is firefox_desktop and gecko which don't use push-mode?
There was a problem hiding this comment.
they use update mode, which is essentially push mode with daily batches, so they would still be impacted.
There was a problem hiding this comment.
This is basically the same issue Jan-Erik ran into with this bug: https://bugzilla.mozilla.org/show_bug.cgi?id=1815432
| - netwerk/metrics.yaml | ||
| - netwerk/protocol/http/metrics.yaml | ||
| - toolkit/components/cookiebanners/metrics.yaml | ||
| - toolkit/components/crashes/metrics.yaml |
There was a problem hiding this comment.
As I understand it, due to push mode the results of historical scans will remain in place from metrics|ping|tag files that are no longer referenced in repositories.yaml, so removing these from gecko should be fine.
But, adding these files to an application that depends on gecko would cause duplicated metric identifiers.
After doing some work to fix duplicate metric checks, I was able to produce the following email message in testing:
New Email
From: telemetry-alerts@mozilla.com
To: crash-reporting-wg@mozilla.org,chutten@mozilla.com,stability@mozilla.org,glean-team@mozilla.com
Subject: Glean: Duplicate metric identifiers detected
Body:
Glean has detected duplicated metric identifiers coming from the product 'firefox-desktop'.
- 'crash.process_type' defined more than once in firefox-desktop, gecko
- 'crash.startup' defined more than once in firefox-desktop, gecko
- 'crash.time' defined more than once in firefox-desktop, gecko
- 'crash.uptime' defined more than once in firefox-desktop, gecko
What to do about this:
1. File a bug to track your investigation. You can just copy this email into the bug Description to get you started.
2. Reply-All to this email to let the list know that you are investigating. Include the bug number so we can help out.
3. Rename the most recently added metric to be more specific. See [1]
4. Make sure a Glean team member reviews any patches. Care needs to be taken that the resolution of this problem is schema-compatible.
If you have any problems, please ask for help on the #glean Slack channel. We'll give you a hand.
What this is:
We have a system called probe-scraper [2] that scrapes the metric information from all Mozilla products using the Glean SDK. All the scraped data is available on the probeinfo service [3]. The scraped definition is used to build things such as the probe-dictionary [4] and other data tools. It detected that one metric that was recently added has an identifier collision with some metric that already existed in the application namespace. So it sent this email out, encouraging you to fix the problem.
What happens if you don't fix this:
The metrics will compete to send their data in pings, making the data unreliable at best.
You can do this!
Your Friendly, Neighborhood Glean Team
[1] - https://mozilla.github.io/glean/book/user/adding-new-metrics.html#naming-things
[2] - https://github.com/mozilla/probe-scraper
[3] - https://probeinfo.telemetry.mozilla.org/
[4] - https://telemetry.mozilla.org/probe-dictionary/
Leave crash-related things alone (See mozilla#551). And while we're here, remove the note about the now-defunct `pine`.
Leave crash-related things alone (See #551). And while we're here, remove the note about the now-defunct `pine`.
|
per https://bugzilla.mozilla.org/show_bug.cgi?id=1815432#c3 i'm going to merge this and run probe scraper in non-update mode to effectively move crash ping from gecko to firefox-desktop, deleting the (empty) glean crash ping tables for all other impacted apps. |
4 participants
This (automated) patch updates the list from metrics_index.py.
For reviewers:
The source code of this automation bot lives in https://github.com/mozilla/probe-scraper/tree/main/fog-updater.