1291445 - Record the time of ingestion for completed jobs

Status: RESOLVED WONTFIX

(Tree Management :: Treeherder: Data Ingestion, defect)

Description

[Kyle Lahnakoski [:ekyle]]

Treeherder shows jobs in grey while it waits for the job to complete.  Please record the time that the job first moves from this pending state to a final state.

This would be nice to have in the case of measuring end-to-end times:  Treeherder's in an important role, where a job has not "completed" until it shows in Treeherder as such.  The time between a job (or task) being technically completed to the time Treeherder shows that reality is non-trivial, and should be measured.

Comment 1

[Ed Morley [:emorley]]

If I read this correctly you mean something like "Record the ingestion time of completed jobs" aka "Provide a way to calculate the lag between a job finishing and it being displayed as such in Treeherder".

We already have last_modified, however I believe by design this gets updated when someone classifies a failure, and not just when the job state was updated.

We could either add a new field on the jobs table (although this is pretty laborious, given the size of the tables), or as a workaround in the meantime, if you only counted jobs with no classification, I think the last modified will actually be the final "job marked as completed" time (though Cameron or someone else will be able to confirm).

Comment 2

[Armen [:armenzg]]

11:56 AM <armenzg> ekyle: where does this bug fit? is it still a blocker? https://bugzilla.mozilla.org/show_bug.cgi?id=1291445
11:57 AM <ekyle> yes, the TH ingestion time does block the end-to-end time meta bug (whatever that one is)
11:58 AM <ekyle> armenzg: but the bug is not a blocker for the visualization:  The visualization will grow independently of all the tracked steps
11:58 AM <ekyle> armenzg: the TH ingestion time will just be another step
11:59 AM <armenzg> ekyle: got ya, I will block it to the "thunder-try" main tracking bug

Comment 3

[Ed Morley [:emorley]]

For jobs with no classifications, the time of ingestion is already available as the last_modified field on the jobs table. Note that this doesn't necessarily mean the time the user will see the job, if they already had the treeherder tab open, since they'll have to wait for the next UI poll. Also, that's just the job, there's then the time to parse the log and generate bug suggestions.

However aside from one-off manually timing or by papertrail log inspection, there's not much we can do here, particularly given our current headcount shortage.