Hello,
my cluster is currently showing a metadata imbalance. Normally, all OSDs
have around 23GB metadata (META column), but 4 OSDs out of 56 have 34 GB
metadata. Compacting reduces the data for some OSDs, but not for others.
OSDs where the compaction worked quickly grow to back to 34GB.
Our cluster configuration:
* 8 nodes, each with 6 HDDs OSDs and 1 SSD used for blockdb and WAL
* k=4 m=2 EC
* v14.2.14
Normal OSD:
ID CLASS WEIGHT REWEIGHT SIZE RAW USE DATA OMAP META
AVAIL %USE VAR PGS STATUS
40 hdd 11.09470 1.00000 11 TiB 8.6 TiB 8.4 TiB 1.3 GiB 23 GiB 2.5
TiB 77.15 1.01 130 up
Big OSD:
ID CLASS WEIGHT REWEIGHT SIZE RAW USE DATA OMAP META
AVAIL %USE VAR PGS STATUS
0 hdd 11.09499 1.00000 11 TiB 8.6 TiB 8.4 TiB 1.8 GiB 30 GiB 2.5
TiB 77.59 1.02 130 up
There are 56 OSDs in the cluster, 4 OSDs of which are bigger. These OSDs
are all in different hosts.
Why is that? Is that dangerous or could lead to problems such as
performance degrades?
Thanks,
Paul
Show replies by date