ZBX-11439 is similar/related.

ZBX-11412 may be related, too.

ZBX-11454 was created to deal with the fallout while this issue will be kept open to fix the housekeeper.

ZBX-11768 also may be related.

Fixed in svn://svn.zabbix.com/branches/dev/ZBX-11426

Modified the filters in the housekeeping_events() function for checking through a subquery if an event have an associated problem in the problem table. Remove only the events without a corresponding record (open or closed).
Also, reordered the deletion query for an easier adding of the filter.

Successfully tested, please review minor changes in r72783

abs Looks good. CLOSED

Released in:

• pre-3.2.9rc1 r72945-r72945
• pre-3.4.3rc1 r72947
• pre-4.0.0alpha1 (trunk) r72948

this might be worth documenting in the housekeeper section (and maybe also in the upgrade notes for 3.2.9 and 3.4.3)

richlv

Maybe a good idea, as now the housekeeper behaviour is explicit regarding how some types of events are kept or deleted. The issue itself was already mitigated in the past through another task and this is just the completion of that work.

indeed, currently the behaviour seems to be completely undocumented

With event housekeeping period set to 1d (or close to it) there is a danger of recovery events being removed while the recovered events are still in problem table.

I'm not sure if it's worth adding more complexity to event deleting queries (although it would be the safest way). I think acceptable workaround would be to call housekeeping_problems() before housekeeping_events() function. As the event housekeeping period cannot be less than problem cleanup period (24h) this would ensure that problems are removed from problem table before corresponding events are removed from events table.

wiper So it was decided to have proper fix. To do it we need to add problem.r_eventid index and also check for r_eventid when removing events.
Still it's better to swap housekeeping_problems() and housekeeping_events() calls so problems table could have potentially less records when housekeeping_events() is called.

Fixed in svn://svn.zabbix.com/branches/dev/ZBX-11426

Swap the calls to housekeeping_problems() and housekeeping_events().
Logically, it's safer to remove old problems first and after that the
related events if necessary.

Also added a filter to the events delete queries for checking the
problem.r_eventid field.
In this way we ensure that any event that is associated with a problem
record in any way (being it a problem or recovery event), will not be
deleted before the problem record itself, but only after.

Successfully tested, please review coding style fixes in r74679

abs style fix ok. CLOSED

Released in:

• pre-3.2.11rc1 r74716
• pre-3.4.5rc1 r74717
• pre 4.0.0alpha1 (trunk) r74718

martins-v

The housekeeper final behaviour after this change is that an event
will be deleted ONLY if is not associated with a problem in any way.
This mean that if an event is either a PROBLEM or RECOVERY event,
it will not be deleted until the related problem record is removed.

Also, now the housekeeper will delete problems first and events
after, for avoiding potential problems with stale events or problem
records.