New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 731773 link

Starred by 1 user

Issue metadata

Status: Fixed
Owner:
Closed: Jun 2018
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 3
Type: Bug



Sign in to add a comment

Triaged alerts disappear/can't be found.

Project Member Reported by erikc...@chromium.org, Jun 9 2017

Issue description

I recently triaged a group of 100+ alerts to be marked as ignored:

https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgprifogkM

[The increase was from turning on native heap profiling on this benchmark].

I then tried to see triaged alerts, but to my surprise, didn't see any group of 100+ alerts. 

https://chromeperf.appspot.com/alerts?sortby=end_revision&sortdirection=down&sheriff=Desktop%20Memory%20Sheriff&triaged=true

Is the clustering of alerts not stable between triaged/untriaged? I think that maybe all the alerts got unclustered :(
 
Components: -Speed>Telemetry Speed>Dashboard
Owner: benjhayden@chromium.org
Ben, since you're looking at alerts can you investigate? Might be some similarities with https://github.com/catapult-project/catapult/issues/3625
I'm not sure, but it looks like the backend deliberately ungroups ignored alerts:

"The bug has been marked invalid/ignored. Kick it out of the group."
https://github.com/catapult-project/catapult/blob/master/dashboard/dashboard/models/alert_group.py#L149

Am I reading it right?

Maybe instead of kicking the alert out of the group, it should split the group into an ignored group and an untriaged group?

Nice find, hmm yeah I'm not clear on why it does that, maybe Annie knows?

I think you were looking at whether it was possible to move this entirely out of the backend and make grouping simply a client-side concept? If that's possible, this won't be much of an issue anymore.
I wasn't around when this was implemented, but I think the idea is that we don't want to show invalid alerts along with the valid ones in the same group because it could confuse people trying to figure out which graphs are important.

I think splitting into a second group as per #2 is the right thing to do.
Status: Fixed (was: Untriaged)
Grouping is done in the frontend now, so it's much easier to experiment with this algorithm. I just changed V2SPA to ungroup alerts with different bug_ids per #4. I'll make a note to ask sheriffs to keep an eye on the grouping algorithm when dogfooding V2SPA. I'll close this, but please feel free to email me directly or use the "File a bug" button with any feedback.

Sign in to add a comment