1171129 - xperf .etl files should be uploaded even when the test fails

Status: RESOLVED WONTFIX

(Testing :: Talos, defect)

Description

[(No longer employed by Mozilla) Aaron Klotz]

This is important because we sometimes need to look at the raw data to determine what's going on. This is particularly important for diagnosing failures.

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

a good bug to help out!

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

this is funny because we only upload the .etl on failure!  quite possible we have the wrong type of failure.

to upload a file all we need to do is copy it to the MOZ_UPLOAD_DIR, and we do that for the raw .etl file:
http://hg.mozilla.org/build/talos/file/0600cd6a8eba/talos/xtalos/etlparser.py#l460

on treeherder you can click on a Talos 'x' job, then in the preview pane, click the 'Job Details' to see the link.  Unfortunately for talos e10s, this is not the case:
https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&filter-searchStr=xperf&exclusion_profile=false

Why are we getting this- I think this is because we have 2 parts to xperf:
1) mainthreadIO
2) xperf fileIO/networkIO

I believe we find errors in mainthreadIO first and never upload- we should optimize that.  Looking in ttest.py where we run the steps:
http://hg.mozilla.org/build/talos/file/0600cd6a8eba/talos/ttest.py#l559

we parse the mainthread first, then do the cleanup step, and the cleanup step is where we do the .etl parsing.

Most likely the solution here will be to do the cleanup step if we have a mainthread failure.

Comment 3

[Joel Maher ( :jmaher ) (UTC -8)]

closing out old bugs that haven't been a priority