Use of PerformanceCounter

[Rogan Carr <ro...@gmail.com>]

Hi All,

PerformanceCounter, used in the MachineStatus class, is not CoreCLR
compatible, so I've been investigating updating it to something that can
work cross-platform.

In my investigations, I noticed that I am never actually able to get the
machine status from the clusters in the current implementation. I always
get a permissions error. [1] I am running on Azure HDInsight clusters
(Windows, Hadoop 2.7.0).

Does anybody have an application or cluster setup where they do see
performance numbers in their logs?

I'm wondering if we can just remove the current implementation, leaving a
stubbed out class, and create a new JIRA asking for an implementation.

Best,
Rogan

[1] The permissions error
"""
WARNING: ExceptionCaught UnauthorizedAccessException Cannot obtain machine
status due to error
Encountered error [System.UnauthorizedAccessException: Access to the
registry key 'Global' is denied.
"""

Re: Use of PerformanceCounter

[Rogan Carr <ro...@gmail.com>]

Hi Markus,

I'll start with an #ifdef solution and track it here:
https://issues.apache.org/jira/browse/REEF-1817 (I had a generic "Fix
issues in REEF.Common that I renamed for this purpose.)

Longer term, it looks like there's no cross-platform APIs for measuring
system performance yet in .NET Standard 2.0. My preference would be to file
a JIRA and come back to it when it's either pressing and/or there is a
supported methodology for measuring performance.

Best,
Rogan



On Tue, Jul 18, 2017 at 6:18 PM, Markus Weimer <ma...@weimo.de> wrote:

> On Tue, Jul 18, 2017 at 5:02 PM, Rogan Carr <ro...@gmail.com> wrote:
> > I'm wondering if we can just remove the current implementation, leaving a
> > stubbed out class, and create a new JIRA asking for an implementation.
>
> We could #ifdef it and just report -1 for all values when on CoreCLR
> for now? Needless to say, we should have a JIRA to fix that :)
>
> Markus
>

Re: Use of PerformanceCounter

[Markus Weimer <ma...@weimo.de>]

On Tue, Jul 18, 2017 at 5:02 PM, Rogan Carr <ro...@gmail.com> wrote:
> I'm wondering if we can just remove the current implementation, leaving a
> stubbed out class, and create a new JIRA asking for an implementation.

We could #ifdef it and just report -1 for all values when on CoreCLR
for now? Needless to say, we should have a JIRA to fix that :)

Markus