New issue
Advanced search Search tips

Issue 783259 link

Starred by 1 user

Issue metadata

Status: Fixed
Owner:
Closed: Nov 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 0
Type: Bug



Sign in to add a comment

Catapult Autoroller is broken

Project Member Reported by eakuefner@chromium.org, Nov 9 2017

Issue description

The Catapult autoroller is in a weird state:

* It says that the most recent roll https://chromium-review.googlesource.com/c/chromium/src/+/756964 failed, but it actually landed.
* The full history shows this CL https://chromium-review.googlesource.com/c/chromium/src/+/757817 with an additional Catapult commit, but this roll is not displayed in the list on catapult-roll.skia.org
* There's more than a day worth of Catapult commits to pick up after f83c79, which is the revision of the most recent roll, but the roller is sitting idle.
* Clicking the stop button seems to hang and not actually stop the roller.
 

Comment 1 by bore...@google.com, Nov 9 2017

Status: Started (was: Untriaged)
Looking into this now.

#1 is because the roll was re-opened and landed after the roller closed it.  The roller does not continue to poll old CLs once they are closed.

Comment 2 by bore...@google.com, Nov 9 2017

Looks like "git cl upload" was the last thing to run, on 11/7/2017 at 6:59 pm EST.  That matches the CL from #2.  There are no errors in the logs, so it's not clear what happened here.  Based on the CL (https://chromium-review.googlesource.com/c/chromium/src/+/757817), the CQ bit was never set, which indicates to me that "git cl upload" did not finish.  If that's the case, the fact that a sub-process was allowed to hang for two days is a big problem.

In the meantime, I restarted the server and it seems to be working again.

Comment 3 by bore...@google.com, Nov 9 2017

I'm re-enabling the roller now, to see how it behaves.

Comment 4 by bore...@google.com, Nov 9 2017

Okay, I'm still not sure exactly what happened here.  We didn't get alerts because a previous change of mine did not get pushed to prod (user error).  In the future we'll be alerted when any roller does not complete a cycle for 10 minutes.  I'm adding a timeout for "git cl upload" in case that was in fact the cause.
Project Member

Comment 5 by bugdroid1@chromium.org, Nov 9 2017

The following revision refers to this bug:
  https://skia.googlesource.com/buildbot/+/03988495131986efc96da2b290db1d1604526699

commit 03988495131986efc96da2b290db1d1604526699
Author: Eric Boren <borenet@google.com>
Date: Thu Nov 09 19:11:08 2017

[autoroll] Add timeout to "git cl upload"

Bug:  783259 
Change-Id: I932d05db952110c832f1f77acdfb5189e1dc1b88
Reviewed-on: https://skia-review.googlesource.com/69560
Reviewed-by: Ravi Mistry <rmistry@google.com>
Commit-Queue: Eric Boren <borenet@google.com>

[modify] https://crrev.com/03988495131986efc96da2b290db1d1604526699/autoroll/go/repo_manager/deps_repo_manager.go

Woohoo, looks like the Catapult roller picked up the right roll. Thanks Eric for your extremely prompt attention to this issue!

Comment 7 by bore...@google.com, Nov 10 2017

Status: Fixed (was: Started)
No problem!  Looks like we're in good shape now.

Sign in to add a comment