search

redhat-developer / osd-monitor-poc

8 stars 20 forks source link

need pcp2zabbix with PMNS_CHANGED / optional-metric tolerance #37

Closed fche closed 5 years ago

fche commented 5 years ago

https://github.com/performancecopilot/pcp/issues/521

cc: @aditya-konarde

goodwinos commented 5 years ago

There was some discussion a while back with Marko (pmrep author) and other PCP developers around this proposal and the consensus is that it would be too difficult/intrusive to retrofit this feature into pmrep (which is the back-end for pcp2zabbix). pmrep was never designed to re-parse it's metric arguments and config after starting. The work-around is to restart pcp2zabbix, at which time pmrep will re-evaluate the PMNS. The Zabbix receiver would need to tolerate reconnection of the datasource - not sure if that is already the case, but likely.

fche commented 5 years ago

The need continues to exist. At the very least pmrep could be configurable so that it can exit when it receives a PMCD_NAMES_CHANGE. That would give the caller an opportunity to restart it.

fche commented 5 years ago

If the above were combined with tolerance by pmrep to unresolvable metric names (so it doesn't quit, just ignores them), we can have what we need without major pmrep surgery.

goodwinos commented 5 years ago

Thanks to Nathan and Marko, pmrep and all pmrep/pmconfig based pcp2xxx tools in upstream pcp-4.3.0 (including pcp2zabbix) now have support for (a) --ignore-unknown to tolerate unresolved metric names and (b) --names-change where supported options are 'ignore' (default), or 'abort'.

See upstream commits : 2aa0081f24cf1db0172a 86e6b73f5aca768a 71d684326aeb22b43 7673438cd7cd8e c1caf330fbd86236 04fd0569bea5c6dc22b and related QA.

These commits should be sufficient for this issue so pcp2zabbix can abort on PMCD_NAMES_CHANGE (and be restarted e.g. by pmmgr), and tolerate unresolvable metric names on startup instead of exiting.

goodwinos commented 5 years ago

As per the above, this feature is now in upstream master and will release with pcp-4.3.0 (December 21).