Re: Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.hpp, line 97
> > 
> >
> > linux/cgroups has an internal {{Result cgroup(pid_t pid, const 
> > string& subsystem)}} that is wrapped for cpu, cpuacct, etc. Can you add an 
> > wrapper for the perf_event cgroup and use that? Yes, it may be less 
> > efficient than looking at the hashmap, but it'll reuse the code and keep 
> > consistency.

We need to cache that value before sampling the events, we can't call cgroup() 
each time when we lookup the pid, because cgroup() always reads from the cgroup 
tasks file.


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, lines 472-482
> > 
> >
> > remove this, see earlier comment.

Removed after switching to pid -> cgroup mapping.


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, line 486
> > 
> >
> > Suggest that you use a hashmap, i.e., just directly 
> > store the cgroup for each pid. Slight tradeoff in size but then you can 
> > avoid findCgroupByPid() and simplify the code.

Done. But note that with pid namespace enabled we could have duplicated pid's 
from different containers. I just add a TODO there.


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, line 502
> > 
> >
> > You don't actually use this as a map? It would be much clearer to use a 
> > vector and explicitly sort with a comparision function on the timestamp. 
> > Suggest just calling it events.
> > 
> > ```
> >   events.push_back(event);
> > }
> > 
> > std::sort(begin(events), end(events), [](...){});
> > ```

I don't see any advantage of sorting it to just using std::map, therefore I 
keep as it is.


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, line 547
> > 
> >
> > Are states enumerated somewhere?

Ideally kernel should export these values, but it doesn't. We could enumerate 
by ourselves, but here we only care about RUNNING which is 0, so...


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, line 560
> > 
> >
> > sort the latencies in schedLatency here?

I don't see any advantage to sort them here than later.


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, lines 625-627
> > 
> >
> > How expensive is this...?

Depends on the workload, we could have tens of thousands events on a busy 
system.


> On Nov. 18, 2015, 12:51 a.m., Ian Downes wrote:
> > src/slave/containerizer/isolators/cgroups/perf_event.cpp, lines 629-632
> > 
> >
> > Users have different kernels, code should determine the version at run 
> > time and act accordingly.

That means we would need two separated code to handle two kernel versions.


- Cong


---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review106958
---


On Sept. 30, 2015, 12:15 a.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated Sept. 30, 2015, 12:15 a.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto 4a16be1f570769f3ce42a50a9da9f4fb1c227999 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 03035dfbfb84470ba39ed9ecfd1eb73890e3f784 
>   src/slave/flags.hpp f76f0f6a0c7a01c7f061a225d7f6ef52be0ee7b5 
>   src/slave/flags.cpp 029aa1eb00e1aa3e92c2155925022c17fd905862 
> 
> Diff: 

Re: Review Request 38074: Calculate schedule latency with trace events

[Ian Downes]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review106958
---



include/mesos/mesos.proto (line 907)


s/percentile/percentiles/



src/slave/containerizer/isolators/cgroups/perf_event.hpp (line 97)


linux/cgroups has an internal {{Result cgroup(pid_t pid, const 
string& subsystem)}} that is wrapped for cpu, cpuacct, etc. Can you add an 
wrapper for the perf_event cgroup and use that? Yes, it may be less efficient 
than looking at the hashmap, but it'll reuse the code and keep consistency.



src/slave/containerizer/isolators/cgroups/perf_event.hpp (line 140)


Format to 80 cols, please.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (lines 116 - 118)


What is this conditional on? It looks like any use of the perf event 
isolator will enable the scheduler latency functionality...



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 130)


const string&



src/slave/containerizer/isolators/cgroups/perf_event.cpp (lines 466 - 476)


remove this, see earlier comment.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 480)


Suggest that you use a hashmap, i.e., just directly store 
the cgroup for each pid. Slight tradeoff in size but then you can avoid 
findCgroupByPid() and simplify the code.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 482)


const string&



src/slave/containerizer/isolators/cgroups/perf_event.cpp (lines 483 - 484)


Please clarify the language, is it "should not" or "does not" :-)



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 485)


const Future<>&



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 486)


Is each Future guaranteed to be ready?



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 487)


Please clarify language, "should" or "is"?



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 489)


Call this variable child?



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 490)


We don't use CHECKs in code unless it truly is an unrecoverable error. 
Please fail the future or act appropriately.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (lines 491 - 492)


No snake_case please, here and everywhere.

Use C++ cast, not C style.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 492)


ditch the intermediate variable and just insert the cast child.get()?

and, using my suggestion above, processes[pid] = cgroup



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 496)


You don't actually use this as a map? It would be much clearer to use a 
vector and explicitly sort with a comparision function on the timestamp. 
Suggest just calling it events.

```
  events.push_back(event);
}

std::sort(begin(events), end(events), [](...){});
```



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 506)


const TraceEvent&



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 514)


It would be nice to have helpers which returned the correct type for event 
fields.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 515)


No CHECKs.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 516)


C++ cast.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 526)


ditto.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 527)


ditto.



src/slave/containerizer/isolators/cgroups/perf_event.cpp (line 529)


ditto.

Re: Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/
---

(Updated Sept. 30, 2015, 12:15 a.m.)


Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.


Changes
---

Cleanup and rebase


Bugs: MESOS-2769
https://issues.apache.org/jira/browse/MESOS-2769


Repository: mesos


Description
---

Finally, calculate schedule latency with the sched trace events, and add it to 
our statistics


Diffs (updated)
-

  include/mesos/mesos.proto 4a16be1f570769f3ce42a50a9da9f4fb1c227999 
  src/slave/containerizer/isolators/cgroups/perf_event.hpp 
1f722ef3ef7ab7fce5542d4affae961d6cec2406 
  src/slave/containerizer/isolators/cgroups/perf_event.cpp 
03035dfbfb84470ba39ed9ecfd1eb73890e3f784 
  src/slave/flags.hpp f76f0f6a0c7a01c7f061a225d7f6ef52be0ee7b5 
  src/slave/flags.cpp 029aa1eb00e1aa3e92c2155925022c17fd905862 

Diff: https://reviews.apache.org/r/38074/diff/


Testing
---

manual tests


Thanks,

Cong Wang

Re: Review Request 38074: Calculate schedule latency with trace events

[Mesos ReviewBot]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review101063
---


Patch looks great!

Reviews applied: [37540, 37541, 38074]

All tests passed.

- Mesos ReviewBot


On Sept. 30, 2015, 12:15 a.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated Sept. 30, 2015, 12:15 a.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto 4a16be1f570769f3ce42a50a9da9f4fb1c227999 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 03035dfbfb84470ba39ed9ecfd1eb73890e3f784 
>   src/slave/flags.hpp f76f0f6a0c7a01c7f061a225d7f6ef52be0ee7b5 
>   src/slave/flags.cpp 029aa1eb00e1aa3e92c2155925022c17fd905862 
> 
> Diff: https://reviews.apache.org/r/38074/diff/
> 
> 
> Testing
> ---
> 
> manual tests
> 
> 
> Thanks,
> 
> Cong Wang
> 
>

Re: Review Request 38074: Calculate schedule latency with trace events

[Mesos ReviewBot]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review99634
---


Patch looks great!

Reviews applied: [37540, 37541, 38074]

All tests passed.

- Mesos ReviewBot


On Sept. 18, 2015, 10:13 p.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated Sept. 18, 2015, 10:13 p.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto 4a16be1f570769f3ce42a50a9da9f4fb1c227999 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 03035dfbfb84470ba39ed9ecfd1eb73890e3f784 
>   src/slave/flags.hpp e31a4183170c3442ac4a15365c229391e7e91480 
>   src/slave/flags.cpp add4196dfd06c0f602ff5ebd39960dc05c4cd11f 
> 
> Diff: https://reviews.apache.org/r/38074/diff/
> 
> 
> Testing
> ---
> 
> manual tests
> 
> 
> Thanks,
> 
> Cong Wang
> 
>

Re: Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/
---

(Updated Sept. 18, 2015, 10:13 p.m.)


Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.


Changes
---

Rebase and cleanup


Bugs: MESOS-2769
https://issues.apache.org/jira/browse/MESOS-2769


Repository: mesos


Description
---

Finally, calculate schedule latency with the sched trace events, and add it to 
our statistics


Diffs (updated)
-

  include/mesos/mesos.proto 4a16be1f570769f3ce42a50a9da9f4fb1c227999 
  src/slave/containerizer/isolators/cgroups/perf_event.hpp 
1f722ef3ef7ab7fce5542d4affae961d6cec2406 
  src/slave/containerizer/isolators/cgroups/perf_event.cpp 
03035dfbfb84470ba39ed9ecfd1eb73890e3f784 
  src/slave/flags.hpp e31a4183170c3442ac4a15365c229391e7e91480 
  src/slave/flags.cpp add4196dfd06c0f602ff5ebd39960dc05c4cd11f 

Diff: https://reviews.apache.org/r/38074/diff/


Testing
---

manual tests


Thanks,

Cong Wang

Re: Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

> On Sept. 5, 2015, 2:31 a.m., Guangya Liu wrote:
> > src/slave/flags.cpp, line 306
> > 
> >
> > Interval between which two values?

Interval between two sched sampling. Same as perf events sampling.


> On Sept. 5, 2015, 2:31 a.m., Guangya Liu wrote:
> > src/slave/flags.cpp, line 307
> > 
> >
> > Does this also affect the event, so what is the difference of this flag 
> > and sched_sample_duration?

You need to read the help message, that one is the duration of sampling.


- Cong


---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review97843
---


On Sept. 4, 2015, 11:16 p.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated Sept. 4, 2015, 11:16 p.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 7dc8b7a99074b74ade019ef4df296780650a2e4e 
>   src/slave/flags.hpp b8335aa585376d27b71897f8cbaefaa12f4b3a5c 
>   src/slave/flags.cpp 7539441c685828027db07173e62a4e5fc1e8b54d 
> 
> Diff: https://reviews.apache.org/r/38074/diff/
> 
> 
> Testing
> ---
> 
> manual tests
> 
> 
> Thanks,
> 
> Cong Wang
> 
>

Re: Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/
---

(Updated Sept. 4, 2015, 11:16 p.m.)


Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.


Changes
---

Rebase


Bugs: MESOS-2769
https://issues.apache.org/jira/browse/MESOS-2769


Repository: mesos


Description
---

Finally, calculate schedule latency with the sched trace events, and add it to 
our statistics


Diffs (updated)
-

  include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
  src/slave/containerizer/isolators/cgroups/perf_event.hpp 
1f722ef3ef7ab7fce5542d4affae961d6cec2406 
  src/slave/containerizer/isolators/cgroups/perf_event.cpp 
7dc8b7a99074b74ade019ef4df296780650a2e4e 
  src/slave/flags.hpp b8335aa585376d27b71897f8cbaefaa12f4b3a5c 
  src/slave/flags.cpp 7539441c685828027db07173e62a4e5fc1e8b54d 

Diff: https://reviews.apache.org/r/38074/diff/


Testing
---

manual tests


Thanks,

Cong Wang

Re: Review Request 38074: Calculate schedule latency with trace events

[Mesos ReviewBot]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review97833
---


Patch looks great!

Reviews applied: [37540, 37541, 38074]

All tests passed.

- Mesos ReviewBot


On Sept. 4, 2015, 11:16 p.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated Sept. 4, 2015, 11:16 p.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 7dc8b7a99074b74ade019ef4df296780650a2e4e 
>   src/slave/flags.hpp b8335aa585376d27b71897f8cbaefaa12f4b3a5c 
>   src/slave/flags.cpp 7539441c685828027db07173e62a4e5fc1e8b54d 
> 
> Diff: https://reviews.apache.org/r/38074/diff/
> 
> 
> Testing
> ---
> 
> manual tests
> 
> 
> Thanks,
> 
> Cong Wang
> 
>

Re: Review Request 38074: Calculate schedule latency with trace events

[Guangya Liu]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review97843
---



include/mesos/mesos.proto (line 915)


Can you please add some commentes here to declear what does p50, p90, p95, 
p99 mean? Thanks.



src/slave/flags.cpp (line 306)


Interval between which two values?



src/slave/flags.cpp (line 307)


Does this also affect the event, so what is the difference of this flag and 
sched_sample_duration?


- Guangya Liu


On 九月 4, 2015, 11:16 p.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated 九月 4, 2015, 11:16 p.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 7dc8b7a99074b74ade019ef4df296780650a2e4e 
>   src/slave/flags.hpp b8335aa585376d27b71897f8cbaefaa12f4b3a5c 
>   src/slave/flags.cpp 7539441c685828027db07173e62a4e5fc1e8b54d 
> 
> Diff: https://reviews.apache.org/r/38074/diff/
> 
> 
> Testing
> ---
> 
> manual tests
> 
> 
> Thanks,
> 
> Cong Wang
> 
>

Re: Review Request 38074: Calculate schedule latency with trace events

[Mesos ReviewBot]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/#review97587
---


Bad patch!

Reviews applied: [37540]

Failed command: ./support/apply-review.sh -n -r 37540

Error:
 2015-09-03 00:58:08 URL:https://reviews.apache.org/r/37540/diff/raw/ 
[12808/12808] -> "37540.patch" [1]
error: patch failed: src/linux/perf.hpp:57
error: src/linux/perf.hpp: patch does not apply
Failed to apply patch

- Mesos ReviewBot


On Sept. 2, 2015, 11:19 p.m., Cong Wang wrote:
> 
> ---
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/38074/
> ---
> 
> (Updated Sept. 2, 2015, 11:19 p.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.
> 
> 
> Bugs: MESOS-2769
> https://issues.apache.org/jira/browse/MESOS-2769
> 
> 
> Repository: mesos
> 
> 
> Description
> ---
> 
> Finally, calculate schedule latency with the sched trace events, and add it 
> to our statistics
> 
> 
> Diffs
> -
> 
>   include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
>   src/slave/containerizer/isolators/cgroups/perf_event.hpp 
> 1f722ef3ef7ab7fce5542d4affae961d6cec2406 
>   src/slave/containerizer/isolators/cgroups/perf_event.cpp 
> 7dc8b7a99074b74ade019ef4df296780650a2e4e 
>   src/slave/flags.hpp e56738e2dfd6593ef8f093687919da287af78f77 
>   src/slave/flags.cpp b36710d6d7a7250bc071a57310a2d54bfb3bc624 
> 
> Diff: https://reviews.apache.org/r/38074/diff/
> 
> 
> Testing
> ---
> 
> manual tests
> 
> 
> Thanks,
> 
> Cong Wang
> 
>

Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/
---

Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.


Bugs: MESOS-2769
https://issues.apache.org/jira/browse/MESOS-2769


Repository: mesos


Description
---

Finally, calculate schedule latency with the sched trace events, and add it to 
our statistics


Diffs
-

  include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
  src/slave/containerizer/isolators/cgroups/perf_event.hpp 
c1578b11ea1afd30929b4ea6f2c8272fb65454ce 
  src/slave/containerizer/isolators/cgroups/perf_event.cpp 
8c3018de8e77a2a00c90559a995eae2b3678e42f 
  src/slave/flags.hpp e56738e2dfd6593ef8f093687919da287af78f77 
  src/slave/flags.cpp b36710d6d7a7250bc071a57310a2d54bfb3bc624 

Diff: https://reviews.apache.org/r/38074/diff/


Testing
---

manual tests


Thanks,

Cong Wang

Re: Review Request 38074: Calculate schedule latency with trace events

[Cong Wang]

---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/38074/
---

(Updated Sept. 2, 2015, 11:18 p.m.)


Review request for mesos, Chi Zhang, Ian Downes, Jie Yu, and Vinod Kone.


Changes
---

Rebase


Bugs: MESOS-2769
https://issues.apache.org/jira/browse/MESOS-2769


Repository: mesos


Description
---

Finally, calculate schedule latency with the sched trace events, and add it to 
our statistics


Diffs (updated)
-

  include/mesos/mesos.proto b1deed4720cab4a89db76a48bc9563bba4f5bf1c 
  src/slave/containerizer/isolators/cgroups/perf_event.hpp 
1f722ef3ef7ab7fce5542d4affae961d6cec2406 
  src/slave/containerizer/isolators/cgroups/perf_event.cpp 
7dc8b7a99074b74ade019ef4df296780650a2e4e 
  src/slave/flags.hpp e56738e2dfd6593ef8f093687919da287af78f77 
  src/slave/flags.cpp b36710d6d7a7250bc071a57310a2d54bfb3bc624 

Diff: https://reviews.apache.org/r/38074/diff/


Testing
---

manual tests


Thanks,

Cong Wang