More instrumentation to diagnose an Android renderer crash.

More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. For
example, this would happen if there was an exec() call.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
Committed: https://crrev.com/aebd42bdab9a59b96e1843c2e9a443e78bd3b2b6
Cr-Commit-Position: refs/heads/master@{#392484}

[Alexei Svitkine (slow), 2016-05-09 19:29:32]

Patchset #1 (id:1) has been deleted

[Alexei Svitkine (slow), 2016-05-09 19:35:13]

Patchset #2 (id:40001) has been deleted

[Alexei Svitkine (slow), 2016-05-09 19:35:19]

Patchset #1 (id:20001) has been deleted

[Alexei Svitkine (slow), 2016-05-09 20:01:04]

Patchset #1 (id:60001) has been deleted

[Alexei Svitkine (slow), 2016-05-09 20:01:22]

Patchset #1 (id:80001) has been deleted

[Alexei Svitkine (slow), 2016-05-09 20:02:26]

Description was changed from

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose this state - for example
an exec() call to nuke the process address space.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

to

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. (For example,
this would happen if there was an exec() call.)

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

[Alexei Svitkine (slow), 2016-05-09 20:02:42]

Description was changed from

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. (For example,
this would happen if there was an exec() call.)

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

to

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. For
example, this would happen if there was an exec() call.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

[Alexei Svitkine (slow), 2016-05-09 20:02:55]

asvitkine@chromium.org changed reviewers:
+ thakis@chromium.org

[Alexei Svitkine (slow), 2016-05-09 20:41:02]

Looking at the base unit test failures.

[Alexei Svitkine (slow), 2016-05-09 20:45:44]

asvitkine@chromium.org changed reviewers:
+ wfh@

[Alexei Svitkine (slow), 2016-05-09 20:45:45]

Okay, base_unnittests should be fixed.

+Will for IPC security review (render_messages.h OWNERS)

Thanks!

[Alexei Svitkine (slow), 2016-05-09 20:47:12]

asvitkine@chromium.org changed reviewers:
+ wfh@chromium.org
- wfh@

[Alexei Svitkine (slow), 2016-05-09 20:47:12]

+Will for IPC security review (render_messages.h OWNERS) for real this time

[Nico, 2016-05-09 20:48:12]

lgtm

https://codereview.chromium.org/1962623005/diff/120001/base/metrics/field_tri...
File base/metrics/field_trial.h (right):

https://codereview.chromium.org/1962623005/diff/120001/base/metrics/field_tri...
base/metrics/field_trial.h:361: // consistent between renderer process
invocation.
invocationS?

[Alexei Svitkine (slow), 2016-05-09 20:50:40]

Thanks!

https://codereview.chromium.org/1962623005/diff/120001/base/metrics/field_tri...
File base/metrics/field_trial.h (right):

https://codereview.chromium.org/1962623005/diff/120001/base/metrics/field_tri...
base/metrics/field_trial.h:361: // consistent between renderer process
invocation.
On 2016/05/09 20:48:11, Nico wrote:
> invocationS?

Done.

[Will Harris, 2016-05-09 21:17:32]

chrome/common/render_messages.h lgtm. how many crashes from this are you
expecting...? :S

[Alexei Svitkine (slow), 2016-05-09 21:26:51]

On 2016/05/09 21:17:32, Will Harris wrote:
> chrome/common/render_messages.h lgtm. how many crashes from this are you
> expecting...? :S

Hopefully, it would just have the existing crashes crash on the new CHECKs to
shed some light on the issue. If the hypothesis doesn't pan out to be true, then
no new crashes. If it does pan out to be true, I guess we'll see how prevalent
the underlying issue is.

[Alexei Svitkine (slow), 2016-05-09 21:26:55]

The CQ bit was checked by asvitkine@chromium.org

[Alexei Svitkine (slow), 2016-05-09 21:26:55]

The patchset sent to the CQ was uploaded after l-g-t-m from thakis@chromium.org
Link to the patchset: https://codereview.chromium.org/1962623005/#ps140001
(title: "Fix typo.")

[commit-bot: I haz the power, 2016-05-09 21:27:25]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1962623005/140001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1962623005/140001

[commit-bot: I haz the power, 2016-05-09 23:02:31]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-05-09 23:02:32]

Try jobs failed on following builders:
  win_chromium_rel_ng on tryserver.chromium.win (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_...)

[Alexei Svitkine (slow), 2016-05-09 23:25:27]

The CQ bit was checked by asvitkine@chromium.org

[commit-bot: I haz the power, 2016-05-09 23:25:47]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1962623005/140001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1962623005/140001

[commit-bot: I haz the power, 2016-05-10 00:15:27]

Description was changed from

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. For
example, this would happen if there was an exec() call.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

to

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. For
example, this would happen if there was an exec() call.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

[commit-bot: I haz the power, 2016-05-10 00:15:28]

Committed patchset #3 (id:140001)

[commit-bot: I haz the power, 2016-05-10 00:16:35]

Description was changed from

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. For
example, this would happen if there was an exec() call.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406
==========

to

==========
More instrumentation to diagnose an Android renderer crash.

We've not observed the instrumentation to detect inconsistent
command-line vs. IPC trial states sent from the FieldTrialList
detecting any issues. One of my next hypothesis on the bug is
something is going to cause us lose both this state and the
field trial session randomized trial dice roll state. For
example, this would happen if there was an exec() call.

To diagnose this, this CL is adding a unique debug token that
is generated during FieldTrialList construction. This is then
tested on the renderer side from OnSetFieldTrialGroup() to
ensure the token is always the same and is also the same as
the one that was used to generate the --force-fieldtrials=
command-line.

BUG=359406

Committed: https://crrev.com/aebd42bdab9a59b96e1843c2e9a443e78bd3b2b6
Cr-Commit-Position: refs/heads/master@{#392484}
==========

[commit-bot: I haz the power, 2016-05-10 00:16:37]

Patchset 3 (id:??) landed as
https://crrev.com/aebd42bdab9a59b96e1843c2e9a443e78bd3b2b6
Cr-Commit-Position: refs/heads/master@{#392484}