This is a bit of a re-design of the way CQ stores attempt cancellation, but is totally worth for a better UI.

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/infra/infra_internal.git/+/cf90bd56eff6c646947a332d18b3b276cf7ebb77

commit cf90bd56eff6c646947a332d18b3b276cf7ebb77
Author: Sergiy Byelozyorov <sergiyb@google.com>
Date: Wed Jun 29 13:19:30 2016

 Issue 628088  has been merged into this issue.

The following revision refers to this bug:
  https://chromium.googlesource.com/infra/infra.git/+/ebcfe3db075b8f7b3cadb2628b37b37fec9bb816

commit ebcfe3db075b8f7b3cadb2628b37b37fec9bb816
Author: tandrii <tandrii@chromium.org>
Date: Tue Jul 26 14:24:12 2016

gerrit_api: add on_behalf_of to set_review method.

R=sergiyb@chromium.org,machenbach@chromium.org
BUG= 618784 

Review-Url: https://codereview.chromium.org/2185653002

[modify] https://crrev.com/ebcfe3db075b8f7b3cadb2628b37b37fec9bb816/infra/libs/gerrit_api/gerrit_api.py
[modify] https://crrev.com/ebcfe3db075b8f7b3cadb2628b37b37fec9bb816/infra/libs/gerrit_api/test/gerrit_api_test.py

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/infra/infra_internal.git/+/088ee397bb3b4f0c25930e04a21f4e5755c1b317

commit 088ee397bb3b4f0c25930e04a21f4e5755c1b317
Author: tandrii <tandrii@google.com>
Date: Tue Jul 26 18:46:39 2016

CL which supports safe removal of CQ label votes of users is ready: https://chromereviews.googleplex.com/480027013

The only thing that isn't really well done is potential race between USER and CQ changing vote at the same time. Most likely it'd be this:
1. User: label-Commit-Queue => 1 (Dry Run)
2. CQ starts and runs dry run, tryjobs are green.
3. User notices green tryjobs decides to trigger full CQ run.
   CQ also notices tryjobs, decides to stop dry run with a success.
4. CQ: label-Commit-Queue => 0 (on behalf of User)
 USER: label-Commit-Queue => 2

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/infra/infra_internal.git/+/aed1569664d3362e68854d27a4b64ce032267b1b

commit aed1569664d3362e68854d27a4b64ce032267b1b
Author: tandrii <tandrii@google.com>
Date: Fri Jul 29 13:55:06 2016

For #11, filed http://crbug.com/632445.

Current state: the CL https://chromereviews.googleplex.com/480027013 with this functionality landed and enables this for infra/infra.

If something goes wrong, land this https://chromereviews.googleplex.com/478067013 to disable the feature for infra/infra.

If all goes well, I'll enable this for all repos.

Ah, forgot to modify infra/infra config: https://chromium.googlesource.com/infra/+/c8d85407f3d7f362e6b606d6476e9f782eb3fc85

OK, things are working well. So, I'll expand it to all projects.

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/infra/infra_internal.git/+/8ece5637ea27c29cf4a0240085d01b47a4b9591d

commit 8ece5637ea27c29cf4a0240085d01b47a4b9591d
Author: tandrii <tandrii@google.com>
Date: Mon Aug 01 13:26:11 2016

And enabled. Let's see now what if anything goes wrong. Metrics: https://app.google.stackdriver.com/monitoring/1029407/tandrii-cq

Fixed configs of boringssl, angle and goma client.

Not sure if part of this bug, but having it clear the CQ bit on behalf of the user is confusing because it’s difficult to audit user behavior vs bot behavior. Especially when tracking down bugs.

Yes, that's true. I think the only way out is finally getting CQ app that would track CQ triggers.

This seems to interact funny with the default CQ setting when a user replies on a change. I'm not sure of the exact trigger, but it seems the default CQ setting is +1? (Which makes sense.) But this means that replying on a CL has a tendency to retry CQ +1 a bunch.

You are right. But, it's easy to change - 1 liner. Shall I do it for BoringSSL?

The change being to undo this change or to switch the default value to 0 instead of +1? Or is there a "default to no change" setting? That one's probably the cleanest; a particular value would bounce around if someone not you is editing labels.

(I'm actually unclear on what the default reply behavior is. It's a little puzzling.)

THIS: "default to no change" setting, so when you reply, labels stay as is.

SGTM. Thanks!

And done: https://boringssl.googlesource.com/All-Projects/+/refs/meta/config

Specifically this commit: https://boringssl.googlesource.com/All-Projects/+/3772877edc6efe01f2fce5e5f6c427552e32a412

OK, this isn't perfect, but works.