Feature #41361
closed
CephInsights - Lacking Sufficient Performance Counters
Added by Alexander Calhoun over 4 years ago.
Updated about 4 years ago.
Description
In order to support performance related RedHat Insights development, it is requested that CephInsights be expands such that OSD/Cluster level performance data is included in the insights data dump.
The a recent insight rule "Bluestore BlueFS Spillover" is blocked due to CephInsights lacking sufficient performance metrics
https://projects.engineering.redhat.com/browse/CEECBA-2463
- Project changed from Ceph to mgr
- Category set to insights module
- Assignee set to Brad Hubbard
Hi Alex,
https://github.com/ceph/ceph/pull/25703 added a warning for bluefs spillover and that could be backported to luminous if required. This warning would be captured by the existing ceph insights module and can be tested for in an insights rule.
It's not clear to me that we want to gather all performance data from all daemons and include that in the insights report. Could you describe what performance data, at a minimum, you would need to accomplish your goal so we can establish a baseline? Please also note that as per https://github.com/RedHatInsights/insights-core/commit/8bd02f188868cb892f289c2bcaabde09272d032a the ceph_insights parser appears to only run on MONs so we may have no access to data that can only be gathered on the OSD hosts themselves (I have no idea at this stage whether that has any bearing but thought it worth mentioning).
Actually, I think the data you need is already captured.
$ jq '.osd_metadata|to_entries[0]|.value.bluefs_db_size' insights.out
"33285996544"
$ jq '.osd_metadata|to_entries[0]|.value.bluestore_bdev_size' insights.out
"4000783007744"
- Status changed from New to Resolved
Also available in: Atom
PDF