Bug #24687
closed
Automatically set expected_num_objects for new pools with >=100 PGs per OSD
Added by Douglas Fuller almost 6 years ago.
Updated over 5 years ago.
Category:
Administration/Usability
Component(RADOS):
Monitor, OSD
Description
Field experience has demonstrated significant performance impact from filestore split and merge activity. The expected_num_objects hint to ceph osd pool create mitigates this impact. Assume that pools created with 50 or more PGs per OSD are expected to store a large number of objects and pre-split their collections automatically.
- Blocked by Bug #24686: change default filestore_merge_threshold to -10 added
- Subject changed from Automatically set expected_num_objects for new pools with >50 PGs per OSD to Automatically set expected_num_objects for new pools with >=50 PGs per OSD
Also include >1024 PGs overall
- Backport changed from luminous to mimic,luminous
- Status changed from In Progress to Fix Under Review
Because a value for expected_num_objects is too difficult to determine automatically, instead we print a suggestion that the user consult pgcalc.
- Subject changed from Automatically set expected_num_objects for new pools with >=50 PGs per OSD to Automatically set expected_num_objects for new pools with >=100 PGs per OSD
- Status changed from Fix Under Review to 7
Removed pgcalc message while pgcalc updates are considered
- Blocks Cleanup #25124: Add message to consult pgcalc for expected_num_objects added
- Status changed from 7 to Pending Backport
- Copied to Backport #25144: mimic: Automatically set expected_num_objects for new pools with >=100 PGs per OSD added
- Copied to Backport #25145: luminous: Automatically set expected_num_objects for new pools with >=100 PGs per OSD added
- Status changed from Pending Backport to Resolved
Also available in: Atom
PDF