Attribute filtering by name patterns (re or glob or..?)

[trws]

Using the current whitelist feature gets us a good bit of the way, but it would be quite useful for some of the flux-core use-cases to be able to specify a whitelist, or blacklist, pattern for attribute names to include rather than the full list of explicit names. It would actually make the need for #6 rather less pressing.

[DavidPoliakoff]

Hey Tom,

You might want to take a look at the work we've been doing in #10 (improved in #13 with your suggestions). The RegexFilter implements what you wanted there, I'll look at integrating it into the Event service as well (which should do what you want)

[trws]

That sounds perfect. Any thoughts on how the overhead would work out using caliper with a filter like that at runtime? I'm actually wondering about runtime switching between a regex filter and a match-none filter so we could just leave it turned on all the time but only create profiles when we want them.

On August 4, 2016 at 4:37:42 PM PDT, David Poliakoff [email protected] wrote:

Hey Tom,

You might want to take a look at the work we've been doing in #10https://github.com/LLNL/Caliper/pull/10 (improved in #13https://github.com/LLNL/Caliper/pull/13 with your suggestions). The RegexFilter implements what you wanted there, I'll look at integrating it into the Event service as well (which should do what you want)

You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/LLNL/Caliper/issues/7#issuecomment-237717553, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AAoStUPj_zh_Ccry5cODG7MeVw_eEai5ks5qcne0gaJpZM4I8CiD.

[DavidPoliakoff]

We haven't benchmarked it yet, but I can't imagine it's anything less than terrible to be honest, it's just the std regex evaluated each time on a begin or end, which makes me a little squeamish. I know @daboehme is looking at a global on/off switch for Caliper, which might be a good stopgap until runtime switching of filters becomes a thing

[trws]

Is the regex compiled every time, or compiled once and reused?

---

Sent from Boxer | http://getboxer.comhttp://bxr.io/PBI3C

On August 6, 2016 at 9:23:43 AM PDT, David Poliakoff [email protected] wrote:

We haven't benchmarked it yet, but I can't imagine it's anything less than terrible to be honest, it's just the std regex evaluated each time on a begin or end, which makes me a little squeamish. I know @daboehmehttps://github.com/daboehme is looking at a global on/off switch for Caliper, which might be a good stopgap until runtime switching of filters becomes a thing

You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/LLNL/Caliper/issues/7#issuecomment-238031673, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AAoStRGwPsGG719Jjjl8XyKBbaLfHEaUks5qdLTrgaJpZM4I8CiD.

[DavidPoliakoff]

Until I get in Monday, it appears to be compiled every time. @davidbeckingsale

[trws]

Easily fixed. If you do, it would probably be good to compile the regex with optimization enabled, chances are the optimization cost will be heavily outweighed by the number of invocations.

On August 6, 2016 at 9:57:57 AM PDT, David Poliakoff [email protected] wrote:

Until I get in Monday, it appears to be compiled every timehttps://github.com/LLNL/Caliper/blob/master/src/common/filters/RegexFilter.h#L24. @davidbeckingsalehttps://github.com/davidbeckingsale

You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/LLNL/Caliper/issues/7#issuecomment-238033483, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AAoStcDraJPOn9jVc0l6JiXEqDSF9vnJks5qdLz8gaJpZM4I8CiD.