Improve add-db-to-osd.sh

Signed-off-by: Anthony D'Atri <[email protected]>

Author: anthonyeleven

add-db-to-osd.sh

@@ -1,15 +1,20 @@
 #!/bin/bash
 # Brett Kelly Oct 2021
+# Anthony D'Atri 2025-04-15
 # 45Drives
-# Version 1.3 stable
+# Version 1.4 stable
 
 usage() { # Help
 cat << EOF
     Usage:	
         [-b] Block DB size. Required. Allowed suffixes K,M,G,T
-        [-d] Device to use as db. Required. Aliased Device name should be used /dev/X-Y
-        [-f] Bypass osd per db warning
-        [-o] OSDs to add db to. Required. Comma separated list of osd.id. <0,1,2,3>
+        [-d] Device to use for DB+WAL. Required. Aliased Device name should be used /dev/X-Y
+        [-f] Bypass OSD per DB warning
+        [-o] OSDs to which to add DB+WAL. Required. Comma separated list of osd.id. <0,1,2,3>
+        [-r] Number of OSDs to share a given WAL+DB offload device, default is 5, which is
+             appropriate for SAS/SATA SSD offload devices.  A value of 10 is usually reasonable
+             for NVMe offload devices, but note that this number of OSDs will fail when the
+             offload device fails.
         [-h] Displays this message
 EOF
     exit 0
@@ -36,25 +41,25 @@ add_lv_tags(){
     lvchange --addtag "ceph.type=db" $DB_LV_DEVICE
 }
 
-check_dependancies(){
-    for i in "${!SCRIPT_DEPENDANCIES[@]}"; do
-        if ! command -v ${SCRIPT_DEPENDANCIES[i]} >/dev/null 2>&1;then
-	        echo "cli utility: ${SCRIPT_DEPENDANCIES[i]} is not installed"
-            echo "jq, and bc are required"
+check_dependencies(){
+    for i in "${!SCRIPT_DEPENDENCIES[@]}"; do
+        if ! command -v ${SCRIPT_DEPENDENCIES[i]} >/dev/null 2>&1;then
+	        echo "The required utility: ${SCRIPT_DEPENDENCIES[i]} is not installed"
+            echo "The jq and bc utilities are required"
 	        exit 1
         fi
     done
 }
 
-# if encountering any error quit, so to not make a mess
+# Quit if we encounter any error, so to not make anything even worse
 set -e
 
-SCRIPT_DEPENDANCIES=(bc jq)
+SCRIPT_DEPENDENCIES=(bc jq)
 FORCE="false"
 PHYSICAL_EXTENT_SIZE_BYTES=4194304
 OSD_PER_DB_LIMIT=5
 
-while getopts 'b:fo:d:h' OPTION; do
+while getopts 'b:fo:d:h:r' OPTION; do
     case ${OPTION} in
     b)
         BLOCK_DB_SIZE=${OPTARG}
@@ -74,6 +79,11 @@ while getopts 'b:fo:d:h' OPTION; do
         OSD_LIST_=${OPTARG}
         IFS=',' read -r -a OSD_LIST <<< "$OSD_LIST_"
         ;;
+    r)
+	OSD_PER_DB_LIMIT=${OPTARG}
+        case $OSD_PER_DB_LIMIT in
+          ''|*[!0-9]*) echo "OSDs per DB device ratio must be an integer" ; exit 1 ;;
+        esac
     h)
         usage
         ;;
@@ -86,10 +96,8 @@ if [ -z $OSD_LIST ] || [ -z $DB_DEVICE ] || [ -z $BLOCK_DB_SIZE_BYTES ]; then
     exit 1
 fi
 
-# If the db device given is a linux sd device then warn if you want to continue
-
-# Check cli depandancies
-check_dependancies
+# Check CLI depandencies
+check_dependencies
 
 BLOCK_DB_SIZE_EXTENTS=$(bc <<< "$BLOCK_DB_SIZE_BYTES/$PHYSICAL_EXTENT_SIZE_BYTES")
 OSD_COUNT="${#OSD_LIST[@]}"
@@ -101,44 +109,44 @@ DB_DEVICE_SIZE_BYTES=$(blockdev --getsize64 $DB_DEVICE)
 # check with wipefs that device has LVM data present
 DB_DEVICE_SIGNATURE=$(wipefs "$DB_DEVICE" --json | jq -r '.signatures | .[0].type // empty')
 # If this is empty the disk is assumed new.
-# If this is LVM2_member the disk is assumed to already have a db lv present it
+# If this is LVM2_member the disk is assumed to already have a DB LV present it
 # If anything else the disk is assumed to have something else on it and should be wiped. Quit with warning
-if [ -z "$LVM_JSON_DEVICE" ] || [ "$DB_DEVICE_SIGNATURE" == "LVM2_member" ];then
+if [ -z "$LVM_JSON_DEVICE" ] || [ "$DB_DEVICE_SIGNATURE" == "LVM2_member" ]; then
     :
 else
-    echo "Disk is not empty nor a LVM device, wipe device first and run again"
+    echo "Device is neither empty nor an LV device. Wipe the device and run again"
     exit 1
 fi
 
-# Get PVS info for the specific disk we want
+# Get PV info for the specific disk we want
 LVM_JSON=$(pvs --units B --nosuffix -o name,vg_name,lv_name,lv_count,lvsize,vg_free --reportformat json ) 
 LVM_JSON_DEVICE=$(echo $LVM_JSON | jq --arg disk "$DB_DEVICE" '.[] |.[].pv | .[] | select(.pv_name==$disk)')
 
-# Check we are using the correct device name
+# Ensure that we are using the correct device
 # if DB_DEVICE_SIGNATURE is LVM2_member and LVM_JSON_DEVICE is empty, then the wrong disk name was used (sd name instead of alias). Quit with warning 
 if [ "$DB_DEVICE_SIGNATURE" == "LVM2_member" ] && [ -z "$LVM_JSON_DEVICE" ];then
-    echo "WARNING: device selected ($DB_DEVICE) has a LVM signature, but could not get LVM info."
-    echo "Wrong disk name was most likely provided, use the device alias name instead of the linux device name"
+    echo "WARNING: device selected ($DB_DEVICE) has an LVM signature, but could not get LVM info."
+    echo "Wrong device name was most likely provided, use the device alias name instead of the Linux device name"
     exit 1
 fi
 
-# are we using an exitsing db device or a new device, if LVM_JSON_DEVICE is empty, and DB_DEVICE_SIGNATURE is empty we have a new disk
+# Are we using an existing DB device or a new device? if LVM_JSON_DEVICE is empty and DB_DEVICE_SIGNATURE is empty we have an empty device
 if [ -z "$LVM_JSON_DEVICE" ] && [ -z "$DB_DEVICE_SIGNATURE" ];then
     DB_VG_NAME="ceph-$(uuidgen)"
 else
-    # if not how do we get db_VG ? inspect from device given
+    # If not how do we get db_VG ? Derive from device given
     DB_VG_NAME="$(echo $LVM_JSON_DEVICE | jq -r '.vg_name' | awk 'NR==1')"
-    # If there is no DB Volume group quit with warning. The disk has a LVM2_memebr signature but no volume group. Wipe disk and run again
+    # If there is no DB Volume group quit with warning. The disk has a LVM2_memebr signature but no volume group. Wipe device and run again.
     if [ -z $DB_VG_NAME ];then
-        echo "WARNING: Device selected ($DB_DEVICE) has a LVM2_member signature, but no volume group"
-        echo "Wipe disk and run again"
+        echo "WARNING: Device selected ($DB_DEVICE) has an LVM2_member signature, but no volume group"
+        echo "Wipe the device and run again"
         exit 1
     fi
-    # Count how many lv dbs are present, add that to input osds and compare to OSD_LIMIT
+    # Count how many LV DBs are present, add that to input OSDs and compare to OSD_LIMIT
     EXISTING_DB_COUNT=$(echo $LVM_JSON_DEVICE | jq -r '.lv_count' | awk 'NR==1')
-    echo "WARNING: device currently has $EXISTING_DB_COUNT db's present"
+    echo "WARNING: device currently has $EXISTING_DB_COUNT dbs present"
     OSD_COUNT=$(bc <<< "${#OSD_LIST[@]}+$EXISTING_DB_COUNT")
-    # set db total device size to the amount of free Bytes in the volume group
+    # set DB total device size to the amount of free Bytes in the volume group
     DB_DEVICE_DISK_SIZE_BYTES=$(echo $LVM_JSON_DEVICE | jq -r '.vg_free' | awk 'NR==1')
 fi
 
@@ -151,16 +159,17 @@ if [ "$FORCE" == "false" ] ; then
     fi
 fi
 
-# Check if total size of db's to be created will fit on db device
+# Check if total size of DBs to be created will fit on DB device
 if [ "$TOTAL_DB_SIZE_BYTES" -gt "$DB_DEVICE_SIZE_BYTES" ] ; then
     echo "Warning: total size of db will not fit on device $DB_DEVICE"
     exit 1
 fi
 
-# Check each osd to see if it present on host
-# Check each osd to see if it already has db device
-# Check current bluestore db size and compare to chosen db size
+# Check each OSD to see if it present on host
+# Check each OSD to see if it already has a DB device
+# Check current BlueStore DB size and compare to supplied DB size
 # Gather ceph-volume output before entering loop as it takes a while to run
+
 CEPH_VOLUME_JSON=$(ceph-volume lvm list --format json)
 for i in "${!OSD_LIST[@]}"; do
     OSD_ID=${OSD_LIST[i]}
@@ -171,7 +180,7 @@ for i in "${!OSD_LIST[@]}"; do
     fi
     DB_CHECK=$(echo $OSD_JSON | jq 'select(.tags["ceph.db_device"])');
     if [ ! -z "$DB_CHECK" ]; then
-        echo "Warning: osd.$OSD_ID already has a db device attached"
+        echo "Warning: osd.$OSD_ID already has a DB device attached"
         exit 1
     fi
     CURRENT_BLOCK_DB_USED_BYTES=$(ceph daemon osd.$OSD_ID perf dump | jq '.bluefs | .db_used_bytes')
@@ -181,9 +190,10 @@ for i in "${!OSD_LIST[@]}"; do
     fi
 done
 
-# Make sure ceph admin keyring is present hs correct permission
+# Make sure the admin keyring is present with correct permissions
 # Remove "set -e" so we can check ceph status error code
-# Then turn it back on after
+# Then turn it back on
+
 set +e
 ceph status > /dev/null 2>&1 ; rc=$?
 set -e
@@ -192,8 +202,7 @@ if [[ "$rc" -ne 0 ]];then
     exit 1
 fi
 
-# If we got this far then all checked are passed
-# Start migration process
+# If we got this far then all checks passed, so start the migration process
 
 if [ -z "$LVM_JSON_DEVICE" ] && [ -z "$DB_DEVICE_SIGNATURE" ];then
     pvcreate $DB_DEVICE
@@ -214,33 +223,41 @@ for i in "${!OSD_LIST[@]}"; do
     chown -h ceph:ceph $DB_LV_DEVICE
     chown -R ceph:ceph $(realpath $DB_LV_DEVICE)
 
-    # Call ceph health check function dont continue unless cluster healthy
+    # Don't continue unless the cluster is healthy
+
     CEPH_STATUS=$(ceph health --format json | jq -r '.status')
     while [ "$CEPH_STATUS" != "HEALTH_OK" ]; do
         echo "Warning: Cluster is not in HEALTH_OK state"
         sleep 2
         CEPH_STATUS=$(ceph health --format json | jq -r '.status')
     done
 
+    OK_TO_STOP=$(ceph osd ok-to-stop $OSD_ID)
+    if [ $OK_TOP_STOP -ne 0 ];
+       echo "Error: stopping osd.$OSD_ID would result in data unavailability"
+       exit 1
+    fi
+
     echo "Set noout"
     ceph osd set noout
     echo "Stop OSD.$OSD_ID"
     systemctl stop ceph-osd@$OSD_ID
+    # Is this a relic of Filestore and thus superfluous?
     echo "Flush OSD Journal"
     ceph-osd -i $OSD_ID --flush-journal
-    echo "Create new db"
+    echo "Create new DB"
     CEPH_ARGS="--bluestore-block-db-size $BLOCK_DB_SIZE_BYTES" ceph-bluestore-tool bluefs-bdev-new-db --path /var/lib/ceph/osd/ceph-$OSD_ID/ --dev-target $DB_LV_DEVICE
-    echo "Migrate old db to new db"
+    echo "Migrate old DB to new DB"
     ceph-bluestore-tool bluefs-bdev-migrate --path /var/lib/ceph/osd/ceph-$OSD_ID/ --devs-source /var/lib/ceph/osd/ceph-$OSD_ID/block --dev-target /var/lib/ceph/osd/ceph-$OSD_ID/block.db
-    echo "Update LV tags on block and db"
+    echo "Update LV tags on block and DB devices"
     add_lv_tags
     echo "unmount OSD.$OSD_ID"
     umount /var/lib/ceph/osd/ceph-$OSD_ID/
     echo "Activate OSD.$OSD_ID"
     ceph-volume lvm activate $OSD_ID $OSD_FSID
     echo "Unset noout"
     ceph osd unset noout
-    echo "Verify osd is back up before continuing"
+    echo "Verify OSD is up before continuing"
     OSD_STATE=$(ceph osd tree --format json | jq --arg id "$OSD_ID" -r '.nodes[] | select(.id == ($id |tonumber)) | .status')
     echo "OSD_STATE:  $OSD_STATE"
     while [ "$OSD_STATE" != "up" ]; do