1289599 - Refresh the bugscache when a bug is successfully filed with the bugfiler tool

Status: RESOLVED WONTFIX

(Tree Management :: Treeherder, defect)

Description

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

Not sure how easy this would be (or if it might overload something by doing this too often), but I think it would be useful for the bugscache to be near-instantly refreshed after a new bug is created via the bugfiler tool.

Not sure whether this should forcibly refresh the whole thing like https://github.com/mozilla/treeherder/blob/01bb61a91be7b7baef664020c23c61fef365c46b/treeherder/etl/bugzilla.py#L22 or if it should just manually create a new bug in the bugscache like https://github.com/mozilla/treeherder/blob/a8717ec2b4e06687a1cec441b79b34527dcabf69/tests/model/test_bugscache.py#L22

Comment 1

[William Lachance (:wlach)]

I think it makes sense to re-fetch the bug from bugzilla, to ensure that the representation in treeherder is consistent with what we have there: but maybe you could restrict the query to *just* the bug that was recently filed. Should be easy enough to do with some minor refactoring of the bugzilla.py file.

Comment 2

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

Would doing this update the suggested bugs for failures that have already been parsed before the bugscache gets updated, or would more steps be needed to include the newly-filed bug?

Use case would be: 
I have a tree (or multiple trees!) full of failed jobs.
I star everything that has a good matching bug.
I still have a tree (or multiple trees!) full of failed jobs that need to be filed.
I file bug 1234 for failure xyz.
Moving on through the list of unstarred failures, I come up to another instance of failure xyz. 
In the failure summary panel, bug 1234 is not in the list of suggested bugs.
Unless I remember each of the 20 bugs I'll end up filing each day, I'm going to end up filing a duplicate bug 1234 for failure xyz.


If inserting bug 1234 into the bugscache immediately after it gets filed means that suggestions for other failures get refreshed automatically, I'll be able to see that I already filed a bug for later instances of failure xyz.

If suggestions for already parsed failures remain with what was already there despite refreshing the bugscache, I'll need to do something else (add duplication detection into the bugfiler itself so it rejects or warns about near-identical bugs, call some function to re-fetch suggestions for failures, etc).

Comment 3

[William Lachance (:wlach)]

From what I understand of the bug suggestions component in treeherder, I think only future jobs will include the new failure, since generating the bug suggestions is done just after log parsing.

I feel like we should probably tackle this in stages:

Step 1: Immediately update the bugscache with the newly filed bug after it's created. Should be relatively straightforward.
Step 2: Update the bug suggestions for previously processed jobs. This might be easier after bug 1281820 is finished.

My recommendation would  we do step (1) now, and then maybe step (2) later? :jgraham, :camd, :emorley may also have ideas here, maybe we could discuss in a treeherder meeting.

Comment 4

[Cameron Dawson [:camd]]

I concur with Will on that.  We only generate the bug suggestions right after log parsing.  And, yeah, step 1 is nice, low-hanging fruit for that.

Off the top of my head, step 2 could be approached in a couple ways:

1. you could request to update the bug suggestions on the selected/pinned jobs manually
2. Have some background process to re-evaluate bug suggestions back to a certain time window (or other window?)

#1 would probably be a bit easier.  Though the downsides are pretty obvious:  What if you don't realize this falls into a case where it should be updated?  Especially if another sheriff was the one who entered the bug.

Perhaps we'd need both?

Comment 5

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

Not to spitball too much on step 2 when step 1's nowhere near started, but maybe we could gather up all individual unstarred jobs in the browser of the person filing the new bug and just re-process them with the updated bugscache? It wouldn't catch identical failures on other trees, but it'd do better than the manual re-processing for #1, while hopefully being a smaller list of things to reprocess than #2.

Comment 6

[William Lachance (:wlach)]

(In reply to Wes Kocher (:KWierso) from comment #5)
> Not to spitball too much on step 2 when step 1's nowhere near started, but
> maybe we could gather up all individual unstarred jobs in the browser of the
> person filing the new bug and just re-process them with the updated
> bugscache? It wouldn't catch identical failures on other trees, but it'd do
> better than the manual re-processing for #1, while hopefully being a smaller
> list of things to reprocess than #2.

Maybe a "refresh bug suggestions" button? Bug 1281820 being fixed would make that a bit more straightforward, I suspect. :)

Comment 7

[Ed Morley [:emorley]]

New bugs are fetched once an hour. Even if that happened immediately, old bug suggestions still wouldn't be updated.

All of this will be simpler in the future when bugs aren't the canonical identifier.