User Tools

Site Tools


storage:ceph

CEPH Repair

ceph health detail
HEALTH_ERR 1 scrub errors; Possible data damage: 1 pg inconsistent
OSD_SCRUB_ERRORS 1 scrub errors
PG_DAMAGED Possible data damage: 1 pg inconsistent
    pg 3.31 is active+clean+inconsistent, acting [5,2,0]

Corrupted PG on OSD 5,2,0

ceph pg repair 3.31
2019-07-29 10:01:54.975649 mon.cloud-gis00 (mon.0) 21584 : cluster [INF] Health check cleared: OSD_SCRUB_ERRORS (was: 1 scrub errors)
2019-07-29 10:01:54.975690 mon.cloud-gis00 (mon.0) 21585 : cluster [INF] Health check cleared: PG_DAMAGED (was: Possible data damage: 1 pg inconsistent)
2019-07-29 10:01:54.975709 mon.cloud-gis00 (mon.0) 21586 : cluster [INF] Cluster is now healthy
2019-07-29 10:01:52.358272 osd.5 (osd.5) 428 : cluster [ERR] 3.31 shard 0 soid 3:8df0528b:::rbd_data.9f8f474b0dc51.0000000000002485:head : candidate had a read error
2019-07-29 10:01:52.358608 osd.5 (osd.5) 429 : cluster [ERR] 3.31 repair 0 missing, 1 inconsistent objects
2019-07-29 10:01:52.358616 osd.5 (osd.5) 430 : cluster [ERR] 3.31 repair 1 errors, 1 fixed

Edit crush-map

ceph osd getcrushmap -o /tmp/crushmap
crushtool -d /tmp/crushmap -o crush_map

crushtool -c crush_map -o /tmp/crushmap
ceph osd setcrushmap -i /tmp/crushmap

Turn cache on

[client]
rbd_cache = true

May improve performance

osd_enable_op_tracker = false
throttler perf counter = false 

Change device class

If the automatic device class detection gets something wrong (e.g., because the device driver is not properly exposing information about the device via /sys/block), you can also adjust device classes from the command line:

$ ceph osd crush rm-device-class osd.2 osd.3
done removing class of osd(s): 2,3
$ ceph osd crush set-device-class ssd osd.2 osd.3
set osd(s) 2,3 to class 'ssd'

Partitions

# types
type 0 osd
type 1 host
type 2 chassis
type 3 rack
type 4 row
type 5 pdu
type 6 pod
type 7 room
type 8 datacenter
type 9 region
type 10 root

OSD Weight

ceph osd crush set 0 0.5 pool=default host=proxmox01
ceph osd crush set 1 0.5 pool=default host=proxmox02
ceph osd crush set 2 0.5 pool=default host=proxmox03

Benchmark

rados -p ceph bench 60 write --no-cleanup

Default object size is 4 MB, and the default number of simulated threads (parallel writes) is 16.
-t (threads)
write / seq / read

Show pool stats

rados -p ceph df

Replication

ceph osd pool set data size 3
ceph osd pool set data min_size 2

For n = 4 nodes each with 1 osd and 1 mon and settings of replica min_size 1 and size 4 three osd can fail, only one mon can fail (the monitor quorum means more than half will survive). 4 + 1 number of monitors is required for two failed monitors (at least one should be external without osd). For 8 monitors (four external monitors) three mon can fail, so even three nodes each with 1 osd and 1 mon can fail. I am not sure that setting of 8 monitors is possible.

For three nodes each with one monitor and osd the only reasonable settings are replica min_size 2 and size 3 or 2. Only one node can fail. If you have an external monitors, if you set min_size to 1 (this is very dangerous) and size to 2 or 1 the 2 nodes can be down. But with one replica (no copy, only original data) you can loose your job very soon.

  • Ensure you have a realistic number of placement groups. We recommend
  • approximately 100 per OSD. E.g., total number of OSDs multiplied by 100
  • divided by the number of replicas (i.e., osd pool default size). So for
  • 10 OSDs and osd pool default size = 4, we'd recommend approximately
  • (100 * 10) / 4 = 250.
storage/ceph.txt · Last modified: 2019/07/29 10:04 by Jan Forman