ceph -w
ceph health detail
ceph osd df
ceph osd find
ceph osd blocked-by
ceph osd pool ls detail
ceph osd pool get rbd all

ceph pg dump | grep pgid
ceph pg pgid
ceph osd primary-affinity 3 1.0
ceph osd map rbd obj

Enable/Disable osd

ceph osd out 0
ceph osd in 0

PG repair

ceph osd map rbd file
ceph pg 0.1a query
ceph pg 0.1a 
ceph pg scrub 0.1a #Checks file exists on OSDs
ceph pg deep-scrub 0.1a #Checks file integrity on OSDs
ceph pg repair 0.1a #Fix problems

Delete osd

ceph osd tree
ceph osd out osd.1
sudo systemctl stop ceph-osd@1.service
ceph osd crush remove osd.1
reph auth del osd.1
ceph osd rm osd.1

osd primary-affinity
ceph pg pgid mark_unfound_lost revert|delete
ceph osd liost osdid --yes-i-really-mean-it

Ceph control

ceph daemon osd.1 config get osd_scrub_min-interval

ceph --admin-daemon socket-file-path command
/var/run/ceph/$cluster-$type.$id.asok
perf dump
config show
dump_historic_ops
scrub_path

Version check

ceph tell mon.* version
ceph tell osd.* version

Updates

Update one mon at the time

/usr/share/ceph-ansible/infrastructure-playbooks/rolling_updates.yaml

Ceph cluster flags

ceph osd set

noup    #Do not auto mark osd as up state
nopdown #Do not auto mark OSD as down state
noout   # Do not remove any osds from crush map. Used when performinig maintenance os parts of the cluster. Prevents crush from auto reblancing the cluster when OSDs are stopped.
noin    # Mons will mark running OSDs with the in state. prevents data from beeing auto allocated to that specific OSD
norecover # Prevents any recovery operations. Used when performing maintenance or a cluster shutdown
nobackfill # Prevents any backfill operation. Used when performing maintenance of a cluster. 
noscrub # No scrubbing operations. Scrubbing has performance impact on a PG on the OSD. If a OSD is too slow it will be marked as down.
nodeep-scrub 
norebalance # Prevents rebalancing from runniung

Crush map

ceph osd getcrushmap -o map.bin
crushtool -d map.bin -o map.txt
crushtool -i map.bin --test --show-mapping --rule=5 --num-rep 3
crushtool -c map.txt -o map.bin
ceph osd setcrushmap -i map.bin

ceph osd crush class ls
ceph osd crush tree
ceph osd crush rule ls

ceph osd pool create fast_ssd 32 32 onssd
ceph pg dump pgs_brief
ceph pg dump pgs_brief | grep ^46 #Pool ID
ceph osd lspools
ceph df

Buckets

ceph osd crush add-bucket default-pool root
ceph osd crush add-bucket rack1 rack
ceph osd crush add-bucket rack2 rack
ceph osd crush add-bucket hosta host
ceph osd crush add-bucket hostb host
ceph osd crush move rack1 root=default-pool
ceph osd crush move rack2 root=default-pool
ceph osd crush move hosta rack=rack1
ceph osd crush move hostb rack=rack2

ceph osd crush tree

OSD tools

ceph osd set-full-ratio 0.97
ceph osd set-nearfull-ratio 0.9
ceph osd dump

ceph osd getmap -o ./map.bin
osdmaptool --print ./map.bin
osdmaptool --export-crush ./crush.bin ./map.bin
crushtool -d ./crush.bin -o ./crush.txt 
crushtool -c ./crush.txt -o crushnew.bin
osdmaptool --import-crush ./crushnew.bin ./map.bin
osdmaptool --test-map-pgs-dump ./map.bin

OSD files

/var/lib/ceph/osd/ceph-1/current/0.1a_head/

Moving an OSD journal to an SSD

ceph osd set noout
systemctl stop ceph-osd@3.service
ceph-osd -i 3 --flush-journal
rm -rf /var/lib/ceph/osd/ceph-3/journal
ln -s /dev/sdc1 /var/lib/ceph/osd/ceph-3/journal
ceph-osd -i 3 --mkjournal
systemctl start ceph-osd@3.service
ceph osd unset noout

Placement gorup calc

Total palcement grous = (OSDs * 100) / Number of replicas
Start between 100 to 200 - Never go above 300+

ceph osd pool get rbd pg_num #Total number of pgs in the pool 
ceph osd pool get rbd pgp_num #Total number of of pgs used for hasing in the pool

Rados

rados -p rbd put file /etc/ceph/ceph.conf

Performance

iostat -x

OS Tuning

systcl net.ipv4.tcp_mem
systcl net.ipv4.tcp_rmem
systcl net.ipv4.tcp_wmem
vm.dirty_background_ratio
vm.dirty_ratio
vm.dirty_background_bytes
vm.dirty_bytes
vm.zone_reclaim_node
vm.swappiness
vm.min_free_kbytes

tuned-adm list
tuned-adm active
tuned-adm profile network-latency
tuned-adm profile network-throughput

Enable jumboframes

cat /sys/block/device/queue/scheduler
noop deadline [cfq]
sudo sh -c "echo deadline" > /sys/block/sdb/queue/scheduler"
Schedulers: 
  noop      # noop elevator does nothing. Turns a disk queue in to a FIFO. Select when back-end storage device can also reorder and merge request. Default inside VMs. Useful for devices suchs as SSDs that respond more quickly to request than they are likely to arrive.
  deadline  # groups queued I/O requests together into a read/write batches. Tries to provide a guaranteed latency for requests and prioritizes read requests over writes. Ceph choice for SATA and SAS drives.
  cfq       # Completely Fair Queueing. Multiple I/O classes and priorities to admin can prioritize process ovetr others when it comes to disk access. Handled by ionice command. 
  blk-mq    # Designed for spinning drives. Designed to handle storage with latenfcies of microseconds and missions of IOPS and large internal paralellism. 

osd_mkfs_options_xfs -f -i  size=2048
use noatime,largeio,inode64,swalloc

Design scaling

No more than six OSD journals per SATA SSD device.
No more than 12 OSD journals per NVMe device.

Warning: When an SSD or NVMe device used ot a host joiurnal fails, every OSD using it to host its journal also becomes unavailable.