resource-agents Spread checks over time with with max slop time

Helps avoid dogpiling services that are checked using the same mechanism by adding per-child and per-test random additional sleep time between checks.

Sep 20 '16 17:09 spaceisntsyntax

I don't feel qualified to review this myself, but I don't know who might be considered maintainer for ldirectord now...

@mcnewton, you have been adding features recently - any opinion?

Oct 24 '16 15:10 krig

I don't think this patch as is has even been running. Last time I checked, perl did not have random(), but rand(). And even that rand() probably behaves differently from what you think you are doing there. Feel free to prove me wrong :-)

Also, did you really see bad effects in real life, or are you trying to solve something that does not really exist, purely based on some theoretical assumptions?

If you have seen bad behaviour in real life, have you any metric before and after? Which metric?

Oct 24 '16 20:10 lge

@krig thanks for the include - not really adding features as much as scratching a few itches :).

I haven't tested, but agree with @lge - I'm not convinced it will work as stands.

And also would be interested in what situations this is useful for. I have never seen service checks being a problem for the realservers, and we run checks much quicker than the default. Real traffic usually drown the service checks out entirely in our case.

Oct 24 '16 23:10 mcnewton

You're right that it's not in use, but yes, it is meant to address a real-world problem.

There are several external services that, for various client-side reason, are backed by the same server-side resource, but which the client must address on a variety of ports. Because we have to test the health for each listening port separately, even though the real servers are identical, the real servers end up under a dog-pile of checks all at the same time every few seconds. If the checks were spread out, even by a few hundred milliseconds, the effect would not be harmful.

Thanks.

On Mon, Oct 24, 2016 at 4:02 PM, Lars Ellenberg [email protected] wrote:

I don't think this patch as is has even been running. Last time I checked, perl did not have random(), but rand(). And even that rand() probably behaves differently from what you think you are doing there. Feel free to prove me wrong :-)

Also, did you really see bad effects in real life, or are you trying to solve something that does not really exist, purely based on some theoretical assumptions?

If you have seen bad behaviour in real life, have you any metric before and after? Which metric?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ClusterLabs/resource-agents/pull/848#issuecomment-255849459, or mute the thread https://github.com/notifications/unsubscribe-auth/AP_H2QEnnzHVngR7Xc0luga1dNQBAEWCks5q3Q7PgaJpZM4KB69s .

Oct 25 '16 18:10 spaceisntsyntax