> On Jan. 10, 2014, 12:02 a.m., Benjamin Hindman wrote:
> > How do you intend for these values to get used? How do they compare with 
> > the names for things exposed on the slaves for statistics?
> 
> Niklas Nielsen wrote:
>     For diagnostics mostly. Even though per-framework statistics are already 
> gathered and exposed, node-wide metrics would be another good indicator of 
> performance issues.
>     
>     The current statistics are mostly internal event counters (XXXX_tasks, 
> XXXX_status_updates, ...) and aggregate resources on masters (mem_total, 
> men_used, ...)
>     This patch and https://reviews.apache.org/r/16631/ adds system metrics 
> (and follow the same naming scheme with "system_XXXX"). I do agree that this 
> is not ideal - I am open to ideas.

Would re-using ResourceStatistics make sense here? If we're going to start 
exposing system metrics in the master/slave, we should at least make sure the 
naming is consistent. (I think this is what benh was referring to: how does 
this compare to the names in ResourceStatistics?).

We should think about how we want to expose system metrics in a more general 
sense (if at all), and how this will tie into third party monitoring tools 
(e.g. MESOS-780).


- Ben


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/16634/#review31492
-----------------------------------------------------------


On Jan. 13, 2014, 10:11 p.m., Niklas Nielsen wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/16634/
> -----------------------------------------------------------
> 
> (Updated Jan. 13, 2014, 10:11 p.m.)
> 
> 
> Review request for mesos, Benjamin Hindman, Ben Mahler, and Vinod Kone.
> 
> 
> Bugs: MESOS-581
>     https://issues.apache.org/jira/browse/MESOS-581
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> This patch expose the number bytes of total and free memory in the
> master and slave stats endpoints.
> Similar to https://reviews.apache.org/r/16631/, the new fields are
> named system_mem_total and system_mem_free to disambiguate the
> aggregate mem_total, mem_used, ... in the master stats endpoint.
> Again, I am open to another naming scheme.
> 
> 
> Diffs
> -----
> 
>   src/master/http.cpp d7cd89f 
>   src/slave/http.cpp 1358810 
> 
> Diff: https://reviews.apache.org/r/16634/diff/
> 
> 
> Testing
> -------
> 
> make check and functional testing of /stats.json endpoints.
> 
> 
> Thanks,
> 
> Niklas Nielsen
> 
>

Reply via email to