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

Issue 699248 link

Starred by 2 users

Issue metadata

Status: Archived
Owner:
Last visit > 30 days ago
Closed: Apr 2018
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Windows
Pri: 2
Type: Bug



Sign in to add a comment

Create a tool for bisecting with swarming/isolate

Project Member Reported by jam@chromium.org, Mar 7 2017

Issue description

It would be awesome if we can have a tool to bisect a test failure using swarming and existing build artifacts from a given waterfall builder.

i.e. pass it a waterfall link, starting & ending build number, and optionally extra build flags. I did this today to see when a failure with PlzNavigate started happening to one browser_test on Mac. I manually did this, and even though it was manual it took me only about 10 minutes which is amazing.

Once it figures out which waterfall build the failure started it in, it can also drill down into all the revisions that went into it and see which caused it.
 
Cc: st...@chromium.org robert...@chromium.org
+stgao, +robertocn

I wonder how different this would be from what FindIt is already doing?

Comment 2 by jam@chromium.org, Mar 7 2017

Cc: chanli@chromium.org
Owner: robert...@chromium.org
We were discussing this a while ago, and I think this is a great idea.

I believe one of the concerns we had was how to keep the intermediate build artifacts (to gain the time advantage of incremental compilation) but I IIUC jam@'s post says that we can do that with isolate.

I'll add this to my queue as I am looking for some work to do and this is right down my alley, but I'll discuss this with stgao@ and chanli@ to see if they have different opinions on this.
To be clear, is this tool using swarming and existing build artifacts to find the first failed build and then drill down to a single revision as culprit? 

We have similar approaches in Findit (and Flake Analyzer). It's just we are using try jobs at revision level.

To find the first failed build,
1. for reliable failures, Findit goes backwards one build by another to check if the failure happened, and it'll stop immediately if the failure didn't happen.
2. for flaky tests, Findit generates flakiness trend and identify the first build algorithmically.
Cc: jam@chromium.org
+jam@

Did you use swarming to both build the revisions after the last pass AND run the test?

Comment 6 by jam@chromium.org, Mar 16 2017

I used existing builds from the main waterfall builder (i.e. by specifying its isolate)
Components: Infra>Client>Chrome
Project Member

Comment 8 by sheriffbot@chromium.org, Apr 26 2018

Status: Archived (was: Unconfirmed)
Issue has not been modified or commented on in the last 365 days, please re-open or file a new bug if this is still an issue.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

Sign in to add a comment