RFE: oc tool for bench, use windowed samples vs. rolled up profiles.

[timothysc]

Currently we the register-tool --name=oc ... ​ gives us a rolled up aggregate profile of the code vs a windowed sampling which can be done via the web-connection.  

This issue is a RFE to request to collect pprof data periodically via the default 30 second window size.

e.g. - collect once a minute over the run, each snapshot takes 30 seconds to run.

/cc @jeremyeder

[jeremyeder]

I think @ofthecurerh might already have a working prototype. @ofthecurerh @ekuric can you guys coordinate on who will do this?

[ekuric]

@jeremyeder @timothysc @ofthecurerh I will work on this, need to include #158 too, are we now just looking into web profile, leaving out cpu/mem one?

[timothysc]

web will yield whatever type of data you are looking for be it cpu or memory. The current profile dump of cpu | memory happens from the start time of the process and can be skewed as a result.

Ideally I would like to see the window in which there was a spike, vs. getting a larger view where that spike is averaged away.

[jeremyeder]

Hmm @atheurer have you any existing examples of driver scripts that
leverage pbench-tool-trigger ? We could say ... when cpu > X then scrape
pprof web endpoint?

On Thu, Mar 10, 2016 at 11:11 AM, Timothy St. Clair <
[email protected]> wrote:

web will yield whatever type of data you are looking for be it cpu or
memory. The current profile dump of cpu | memory happens from the start
time of the process and can be skewed as a result.

Ideally I would like to see the window in which there was a spike, vs.
getting a larger view where that spike is averaged away.

—
Reply to this email directly or view it on GitHub
#186 (comment)
.

-- Jeremy Eder