[GRC] Decouple Render Process CPU Measurement from CU Graph

[GRC] Decouple Process CPU Measurement from CU Graph

Currently GRC's process CPU measurement mechanisms are tightly integrated with the process coordination unit implementation. This CL moves the CPU usage measurement from the resource_coordinator service to the content_browser service.

Abstraction Improvements:
Decoupling process CPU measurement from the CU model allows the coordination units to focus on representing browser state within GRC, rather than updating it. In GRC terms, the migrated render process CPU measurement infrastructure is a "probe" that externally triggers updates the CU graph.

Code Health Improvements:
The measurement decoupling makes the coordination unit graph more easily testable. To that end, coordination unit graph mocks have been created to address the repeating pattern of needing a given coordination unit graph topology for (unit)testing).

R=lpy@chromium.org,zhenw@chromium.org
TBR=oysteine@chromium.org

BUG=691886

[matthalp, 2017-06-29 23:12:12]

Description was changed from

==========
Initalize CL


Initialize CL

BUG=
==========

to

==========
[GRC] Decouple Process CPU Measurement from CU Graph

BUG=
==========

[matthalp, 2017-07-01 17:36:48]

Description was changed from

==========
[GRC] Decouple Process CPU Measurement from CU Graph

BUG=
==========

to

==========
[GRC] Decouple Process CPU Measurement from CU Graph

Currently GRC's process CPU measurement mechanisms are tightly integrated with
the process coordination unit implementation. This CL moves the CPU usage
measurement from the resource_coordinator service to the content_browser
service.

Abstraction Improvements:
Decoupling process CPU measurement from the CU model allows the coordination
units to focus on representing browser state within GRC, rather than updating
it. In GRC terms, the migrated render process CPU measurement infrastructure is
a "probe" that externally triggers updates the CU graph.

Code Health Improvements:
The measurement decoupling makes the coordination unit graph more easily
testable. To that end, coordination unit graph mocks have been created to
address the repeating pattern of needing a given coordination unit graph
topology for (unit)testing).

R=lpy@chromium.org,zhenw@chromium.org
TBR=oysteine@chromium.org

BUG=691886
==========

[matthalp, 2017-07-01 17:36:48]

matthalp@google.com changed reviewers:
+ lpy@chromium.org, oysteine@chromium.org, zhenw@chromium.org

[matthalp, 2017-07-01 17:36:48]

matthalp@google.com changed required reviewers:
+ lpy@chromium.org, zhenw@chromium.org

[matthalp, 2017-07-01 17:37:20]

The CQ bit was checked by matthalp@google.com to run a CQ dry run

[commit-bot: I haz the power, 2017-07-01 17:37:28]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-07-01 17:50:59]

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

[commit-bot: I haz the power, 2017-07-01 17:50:59]

Dry run: Try jobs failed on following builders:
  ios-device-xcode-clang on master.tryserver.chromium.mac (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.mac/builders/ios-device-xcode-...)

[matthalp, 2017-07-01 18:31:56]

The CQ bit was checked by matthalp@google.com to run a CQ dry run

[commit-bot: I haz the power, 2017-07-01 18:31:58]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[matthalp, 2017-07-01 19:11:39]

PTAL when convenient -- not urgent.

https://codereview.chromium.org/2964103002/diff/40001/chrome/browser/chrome_b...
File chrome/browser/chrome_browser_main.cc (right):

https://codereview.chromium.org/2964103002/diff/40001/chrome/browser/chrome_b...
chrome/browser/chrome_browser_main.cc:1957:
resource_coordinator::ResourceCoordinatorRenderProcessProbe::GetInstance()
TODO(matthalp): Guard with feature flag. Also expose a sampling rate parameter
and accompanying flag.

[commit-bot: I haz the power, 2017-07-01 19:37:20]

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

[commit-bot: I haz the power, 2017-07-01 19:37:21]

Dry run: Try jobs failed on following builders:
  mac_chromium_rel_ng on master.tryserver.chromium.mac (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.mac/builders/mac_chromium_rel_...)

[Zhen Wang, 2017-07-05 15:37:51]

Peiyong fixed the set-property-and-propagate part in
https://crrev.com/2964803002/. I think his CL will likely to commit before
yours. You can rebase your CL on his and see if it works for you. Thanks!

https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
File
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.h
(right):

https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.h:32:
void CollectAndSendMeasurements();
Make this private?

https://codereview.chromium.org/2964103002/diff/60001/services/resource_coord...
File services/resource_coordinator/coordination_unit/coordination_unit_impl.h
(right):

https://codereview.chromium.org/2964103002/diff/60001/services/resource_coord...
services/resource_coordinator/coordination_unit/coordination_unit_impl.h:71:
void SetPropertyAndPropagate(mojom::PropertyPtr property) override;
Rebase the CL on lpy's CL for this.

[lpy, 2017-07-07 17:49:10]

https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
File
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.cc
(right):

https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.cc:25:
const int kMeasurementIntervalInSeconds = 5;
I would like to understand the reason we chose 5 seconds, and how much overhead
it will have if we change it to, say 0.1 second.

[matthalp, 2017-07-07 18:04:09]

On 2017/07/07 at 17:49:10, lpy wrote:
>
https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
> File
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.cc
(right):
> 
>
https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
>
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.cc:25:
const int kMeasurementIntervalInSeconds = 5;
> I would like to understand the reason we chose 5 seconds, and how much
overhead it will have if we change it to, say 0.1 second.

Right now it is five seconds to ensure the overhead is minimal until there is a
use case that requires a faster sampling rate. For a comparison, the UMA code
samples every 2 minutes. Ideally, I would also like it to be 100ms as well but
have not investigated the overheads yet. I can look into this further.

Measuring process CPU usage is fairly computationally intensive and not just a
simple system lookup. The CPU measurement code has to open and parse a file to
extract the CPU usage information. The code can be found here:
https://cs.chromium.org/chromium/src/base/process/process_metrics.h?type=cs&q...

[matthalp, 2017-07-07 18:36:48]

On 2017/07/07 at 18:04:09, matthalp wrote:
> On 2017/07/07 at 17:49:10, lpy wrote:
> >
https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
> > File
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.cc
(right):
> > 
> >
https://codereview.chromium.org/2964103002/diff/60001/chrome/browser/resource...
> >
chrome/browser/resource_coordinator/resource_coordinator_render_process_probe.cc:25:
const int kMeasurementIntervalInSeconds = 5;
> > I would like to understand the reason we chose 5 seconds, and how much
overhead it will have if we change it to, say 0.1 second.
> 
> Right now it is five seconds to ensure the overhead is minimal until there is
a use case that requires a faster sampling rate. For a comparison, the UMA code
samples every 2 minutes. Ideally, I would also like it to be 100ms as well but
have not investigated the overheads yet. I can look into this further.
> 
> Measuring process CPU usage is fairly computationally intensive and not just a
simple system lookup. The CPU measurement code has to open and parse a file to
extract the CPU usage information. The code can be found here:
>
https://cs.chromium.org/chromium/src/base/process/process_metrics.h?type=cs&q...

lpy@, some discussion about CPU usage overheads:
https://groups.google.com/a/google.com/d/msg/chat-frontend-eng/wySsyLvPFio/n_...