Parents who are non-active should not be rescheduled #5375
Conversation
dnsmichi
changed the title
| if (!parent->GetEnableActiveChecks()) | ||
| continue; | ||
|
|
||
| if (parent->GetNextCheck() >= now + parent->GetRetryInterval()) { |
There was a problem hiding this comment.
This doesn't feel right. Any explanation why one would use retry_interval here?
There was a problem hiding this comment.
Right now there is no limitation to the amount of reschedules that can be forced to a parent by a failed child check. In some cases with about 20 children for one NRPE parent, it will go through its max_check_attempts within a few seconds.
Conceptually the minimum waiting period between checks is the retry_interval during normal scheduling(thus excluding a manual forced reschedule). Theoretically it is possible to have a lower check_interval than retry_interval, however I doubt anyone does that as it makes little sense.
There was a problem hiding this comment.
Alright. I'd like to see this PR tested in various scenarios then, including test cases and excerpts on the next_check values from the API.
|
Coming back here, I wasn't sure about the logic with the retry_interval. Especially some changes made with 2.9 should help prevent false positive check scheduling inserts. |
dnsmichi
added
bug
|
Sorry about the late reply, sometimes things are a little more clear when life turns good. Thanks for your contribution :) |
|
Note to self: Don't use |
Parents who are non-active should not be rescheduled.
This fix also prevents reschedule storms when there are lots of children for a parent. The parent will now only reschedule when its next check is too far in the future, this is determined by the retry interval of the parent.
fixes #5022