Actions
Bug #61743
openmultisite: excessive polling for metadata sync on some shards
% Done:
0%
Source:
Tags:
multisite
Backport:
pacific quincy reef
Regression:
Yes
Severity:
3 - minor
Reviewed:
Description
on some shards, secondary zones may poll for metadata changes in a tight loop instead of sleeping for the rgw_meta_sync_poll_interval. this appears to be triggered by log trimming. as a workaround, you can restart the syncing radosgws on the secondary zones
Updated by Casey Bodley 10 months ago
- Status changed from New to Fix Under Review
- Pull request ID set to 52400
Updated by Mark Kogan 10 months ago
PR https://github.com/ceph/ceph/pull/52400 fixed 2 repro workloads:
1. creating and deleting buckets
2. adding and deleting users
keeps reproduces in 2 workloads:
1. put-bucket-tagging:
time echo -n {1..99999} | xargs -P$(( $(numactl -N 0 -- nproc) / 1 )) -L1 -d' ' -i sh -c 'echo $1 ; set -x ; nice aws --endpoint-url http://0:8000 s3api put-bucket-tagging --bucket hsbench-1-$(printf "%012d" $1) --tagging "TagSet=[{Key=organization,Value=marketing}]" ; set +x' sh {} ; date
polling rate graphs:
2. Bucket Versioning toggle:
time echo -n {1..99999} | xargs -P$(( $(numactl -N 0 -- nproc) / 1 )) -L1 -d' ' -i sh -c 'echo $1 ; set -x ; nice aws --endpoint-url http://0:8000 s3api put-bucket-versioning --bucket hsbench-1-$(printf "%012d" $1) --versioning-configuration Status=Enabled ; set +x' sh {} ; date time echo -n {1..99999} | xargs -P$(( $(numactl -N 0 -- nproc) / 1 )) -L1 -d' ' -i sh -c 'echo $1 ; set -x ; nice aws --endpoint-url http://0:8000 s3api put-bucket-versioning --bucket hsbench-1-$(printf "%012d" $1) --versioning-configuration Status=Suspended ; set +x' sh {} ; date
polling rate graphs:
Updated by Casey Bodley 9 months ago
- Pull request ID changed from 52400 to 52568
Actions