You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@mesos.apache.org by Andrew Schwartzmeyer <an...@schwartzmeyer.com> on 2017/10/26 16:33:41 UTC
Review Request 63277: Windows: Ported CPU and memory isolator tests.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/63277/
-----------------------------------------------------------
Review request for mesos, Akash Gupta, Jeff Coffler, Jie Yu, John Kordich, Joseph Wu, and Li Li.
Bugs: MESOS-6690
https://issues.apache.org/jira/browse/MESOS-6690
Repository: mesos
Description
-------
These test the limitation and usage reporting of the new Windows
`Cpuset` and `Mem` isolators.
Diffs
-----
src/tests/CMakeLists.txt 386e0473c93d0a993248c7818067071d0c761c76
src/tests/containerizer/cpu_isolator_tests.cpp 846b2e255547a02f5eb0590747edca62bc560ac3
src/tests/containerizer/memory_isolator_tests.cpp b8ea5d35b3a0a4820d9ec4c6d7d916dc6101b570
src/tests/mesos.cpp 9185b5bf2175be5b0f6b6a03a04e9e5445bf22fd
Diff: https://reviews.apache.org/r/63277/diff/1/
Testing
-------
Thanks,
Andrew Schwartzmeyer
Re: Review Request 63277: Windows: Ported CPU and memory isolator
tests.
Posted by Andrew Schwartzmeyer <an...@schwartzmeyer.com>.
> On Oct. 26, 2017, 2:59 p.m., Jeff Coffler wrote:
> > src/tests/containerizer/cpu_isolator_tests.cpp
> > Lines 120 (patched)
> > <https://reviews.apache.org/r/63277/diff/1/?file=1869651#file1869651line120>
> >
> > Given what we learned about quoting, why not just do:
> >
> > "powershell -c while ($true) {}"
> >
> > ?
Actually, this should be `-NoProfile -Command` too, and `"` is fine (needed to avoid interpolation by `cmd.exe`), it's `'` that's troublesome.
> On Oct. 26, 2017, 2:59 p.m., Jeff Coffler wrote:
> > src/tests/containerizer/cpu_isolator_tests.cpp
> > Line 163 (original), 172 (patched)
> > <https://reviews.apache.org/r/63277/diff/1/?file=1869651#file1869651line172>
> >
> > A powershell loop wouldn't do it. `Get-Process` is a step in the right direction, but probably has a small amount of kernel time and a lot of PowerShell time for formatting, etc.
> >
> > What about some sort of tight loop that was essentially exclusively a kernel operation, like mapping and unmapping a memory segment in a loop or something? This could be done in C, avoids PowerShell entirely, and would need to be processed by the kernel. Or creating a shared memory mutex, or anything along those lines.
> >
> > The kernel is supposed to be efficient, so it would be hard to consume Kernel time, but pick something that you know is 100% kernel, and do it in a tight loop. That should get you there.
This needs to be a task that can be launched. I really do not want to have to ship yet another binary just for a test. I can if I have to, but there should be a way to accomplish this outside of that.
> On Oct. 26, 2017, 2:59 p.m., Jeff Coffler wrote:
> > src/tests/containerizer/cpu_isolator_tests.cpp
> > Lines 224 (patched)
> > <https://reviews.apache.org/r/63277/diff/1/?file=1869651#file1869651line224>
> >
> > Ditto
This needs to be wrapped in `"` because of the `|`.
> On Oct. 26, 2017, 2:59 p.m., Jeff Coffler wrote:
> > src/tests/containerizer/cpu_isolator_tests.cpp
> > Line 234 (original), 253 (patched)
> > <https://reviews.apache.org/r/63277/diff/1/?file=1869651#file1869651line253>
> >
> > If you're doing a `sleep` operation, that won't consume system time. The kernel will deschedule you until it's time for you to wake up.
> >
> > You might get there by doing a lot of short sleeps, but that's generally risky and unreliable (depends on kernel timing, speed of implementation, etc).
The `sleep` here applies to the test process, not the executing task.
- Andrew
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/63277/#review189358
-----------------------------------------------------------
On Oct. 26, 2017, 9:33 a.m., Andrew Schwartzmeyer wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/63277/
> -----------------------------------------------------------
>
> (Updated Oct. 26, 2017, 9:33 a.m.)
>
>
> Review request for mesos, Akash Gupta, Jeff Coffler, Jie Yu, John Kordich, Joseph Wu, and Li Li.
>
>
> Bugs: MESOS-6690
> https://issues.apache.org/jira/browse/MESOS-6690
>
>
> Repository: mesos
>
>
> Description
> -------
>
> These test the limitation and usage reporting of the new Windows
> `Cpuset` and `Mem` isolators.
>
>
> Diffs
> -----
>
> src/tests/CMakeLists.txt 386e0473c93d0a993248c7818067071d0c761c76
> src/tests/containerizer/cpu_isolator_tests.cpp 846b2e255547a02f5eb0590747edca62bc560ac3
> src/tests/containerizer/memory_isolator_tests.cpp b8ea5d35b3a0a4820d9ec4c6d7d916dc6101b570
> src/tests/mesos.cpp 9185b5bf2175be5b0f6b6a03a04e9e5445bf22fd
>
>
> Diff: https://reviews.apache.org/r/63277/diff/1/
>
>
> Testing
> -------
>
>
> Thanks,
>
> Andrew Schwartzmeyer
>
>
Re: Review Request 63277: Windows: Ported CPU and memory isolator
tests.
Posted by Jeff Coffler <je...@taltos.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/63277/#review189358
-----------------------------------------------------------
src/tests/containerizer/cpu_isolator_tests.cpp
Lines 120 (patched)
<https://reviews.apache.org/r/63277/#comment266383>
Given what we learned about quoting, why not just do:
"powershell -c while ($true) {}"
?
src/tests/containerizer/cpu_isolator_tests.cpp
Line 163 (original), 172 (patched)
<https://reviews.apache.org/r/63277/#comment266386>
A powershell loop wouldn't do it. `Get-Process` is a step in the right direction, but probably has a small amount of kernel time and a lot of PowerShell time for formatting, etc.
What about some sort of tight loop that was essentially exclusively a kernel operation, like mapping and unmapping a memory segment in a loop or something? This could be done in C, avoids PowerShell entirely, and would need to be processed by the kernel. Or creating a shared memory mutex, or anything along those lines.
The kernel is supposed to be efficient, so it would be hard to consume Kernel time, but pick something that you know is 100% kernel, and do it in a tight loop. That should get you there.
src/tests/containerizer/cpu_isolator_tests.cpp
Lines 224 (patched)
<https://reviews.apache.org/r/63277/#comment266385>
Ditto
src/tests/containerizer/cpu_isolator_tests.cpp
Line 234 (original), 253 (patched)
<https://reviews.apache.org/r/63277/#comment266387>
If you're doing a `sleep` operation, that won't consume system time. The kernel will deschedule you until it's time for you to wake up.
You might get there by doing a lot of short sleeps, but that's generally risky and unreliable (depends on kernel timing, speed of implementation, etc).
- Jeff Coffler
On Oct. 26, 2017, 4:33 p.m., Andrew Schwartzmeyer wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/63277/
> -----------------------------------------------------------
>
> (Updated Oct. 26, 2017, 4:33 p.m.)
>
>
> Review request for mesos, Akash Gupta, Jeff Coffler, Jie Yu, John Kordich, Joseph Wu, and Li Li.
>
>
> Bugs: MESOS-6690
> https://issues.apache.org/jira/browse/MESOS-6690
>
>
> Repository: mesos
>
>
> Description
> -------
>
> These test the limitation and usage reporting of the new Windows
> `Cpuset` and `Mem` isolators.
>
>
> Diffs
> -----
>
> src/tests/CMakeLists.txt 386e0473c93d0a993248c7818067071d0c761c76
> src/tests/containerizer/cpu_isolator_tests.cpp 846b2e255547a02f5eb0590747edca62bc560ac3
> src/tests/containerizer/memory_isolator_tests.cpp b8ea5d35b3a0a4820d9ec4c6d7d916dc6101b570
> src/tests/mesos.cpp 9185b5bf2175be5b0f6b6a03a04e9e5445bf22fd
>
>
> Diff: https://reviews.apache.org/r/63277/diff/1/
>
>
> Testing
> -------
>
>
> Thanks,
>
> Andrew Schwartzmeyer
>
>