1299872 - Errors from task setup shell scripts should produce starrable failure lines in Treeherder

Status: RESOLVED WORKSFORME

(Taskcluster :: General, defect)

Description

[Ryan VanderMeulen [:RyanVM]]

+++ This bug was initially created as a clone of Bug #1299869 +++

/home/worker/workspace/build/src/taskcluster/scripts/builder/build-linux.sh: line 7: /home/worker/scripts/xvfb.sh: No such file or directory

https://treeherder.mozilla.org/#/jobs?repo=autoland&selectedJob=2922384

Not sure what we can do here, but it would be nice if that bug were somehow starrable.

Comment 1

[Greg Arndt [:garndt]]

Really the solution here is to remove that error completely, which needs to be done.  This was fallout from me incorporating the xvfb changes, but not updating that particular image.

Comment 3

[Ryan VanderMeulen [:RyanVM]]

I'm a bit confused. Bug 1299869 tracks fixing the specific permafail that's currently being hit. This bug was about said failure not happening in a TH-parsable way. Why the duping?

Comment 4

[Sebastian Kaspari (:sebastian; :pocmo)]

Oh, sorry for that! I thought both are about the error.

Comment 5

[Dustin J. Mitchell [:dustin] (he/him)]

I think the request here is for all shell errors to have a starrable message.  I'm not sure how we'd accomplish that, since by the time an error is signalled any errors have already been logged to stdout/stderr so it's too late to prefix them.  The closest I can think of is adding a trap that prints "[taskcluster:error] something went wrong" which is just going to get everything starred to a single bug.

Could we log every line prefixed with "[taskcluster:error] .."? then you could have your pick of lines to star?

Comment 6

[Nick Alexander :nalexander [he/him]]

(In reply to Dustin J. Mitchell [:dustin] from comment #5)
> I think the request here is for all shell errors to have a starrable
> message.  I'm not sure how we'd accomplish that, since by the time an error
> is signalled any errors have already been logged to stdout/stderr so it's
> too late to prefix them.  The closest I can think of is adding a trap that
> prints "[taskcluster:error] something went wrong" which is just going to get
> everything starred to a single bug.
> 
> Could we log every line prefixed with "[taskcluster:error] .."? then you
> could have your pick of lines to star?

If I were starring regularly, I would want the "error -> click log -> copy line that is relevant -> add new starrable thing" loop to be fast.

Right now, it's not -- if the *task* doesn't identify "this is starrable" (by printing [taskcluster:error]), it's either impossible or very slow to "add new starrable thing".  So, if I were starring regularly, I'd be trying to get out of "task identifies things" and into "make it fast for me to identify things and train the *system* to identify those things for me".  One way to do that would be to tighten the "click log -> find starrable thing -> system identifies those stars without tasks involvement" UX, but there are others.  (Guided learning rather than machine learning, if you will.)

Comment 7

[Greg Arndt [:garndt]]

There are various issues that can happen during the task life cycle that now have star-able errors.  I would take the leap suggesting that this is fixed as far as taskcluster is concerned.