Compacting omaps on OSDs

osd_omap_compaction.sh

#!/bin/bash

# Name: osd_omap_compaction.sh
#
# This will compact the omap dbs on OSDs

# Wait until the cluster is Healthy
HEALTH_OK=${1:-"HEALTH_OK"}
while [[ "$(ceph health)" != "$HEALTH_OK" ]]; do
  sleep 10
done

osds=$(ceph osd crush ls $(hostname -s) | cut -d. -f2)

ceph osd set noout
# For each OSD, we want to run it's commands in the background so all OSDs will run at the same time
for osd in $osds; do
  osd_type=$(ceph osd metadata $osd | grep 'osd_objectstore')
  if echo $osd_type | grep -q filestore; then
    db=$(sudo grep -aEo [[:alpha:]]+db /var/lib/ceph/osd/ceph-${osd}/superblock)
  elif echo $osd_type | grep -q bluestore; then
    db=bluestore-kv
  else
    echo $osd has an invalid type
    exit 1
  fi
  (
    echo "================================================================================"
    sudo systemctl stop ceph-osd@$osd
    omap=/var/lib/ceph/osd/ceph-${osd}/current/omap
    sudo -u ceph ceph-kvstore-tool $db $omap compact
    sudo systemctl start ceph-osd@$osd
  ) &
done
wait

# Wait until all OSDs are back up and in and then increase osd_max_backfills to speed up recovery between nodes
while true; do
  stat=$(ceph osd stat)
  up=$(echo "$stat" | grep -Eo '[[:digit:]]+\s+up' | awk '{print $1}')
  in=$(echo "$stat" | grep -Eo '[[:digit:]]+\s+in' | awk '{print $1}')
  if (( up == in )); then
    ceph tell osd.\* injectargs --osd_max_backfills=2
    ceph osd unset noout
    break
  else
    sleep 10
  fi
done

for osd in $osds; do
  sudo du -sh /var/lib/ceph/osd/ceph-${osd}/current/omap
done