--- /dev/null
+#!/bin/bash
+#
+# ceph-lazy : Be efficient, be lazy !
+#
+# Author: Gregory Charot <gcharot@redhat.com>
+#
+# This is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2.1 of the License, or (at your option) any later version.
+#
+
+# Specify exta option for ceph like the username/keyring/etc. Can also be done with CEPH_ARGS global variable
+#CEPH_OPT="-n client.username"
+VERSION="1.1.2"
+
+#
+# Print info message to stderr
+#
+
+function echoinfo() {
+ printf "INFO: %s\n" "$*" >&2;
+}
+
+
+#
+# Print error message to stderr
+#
+
+function echoerr() {
+ printf "ERROR: %s\n" "$*" >&2;
+}
+
+
+function help() {
+ >&2 echo "Usage : ceph-lazy [-d | -h] [command] [parameters]
+
+Ceph complex quering tool - Version $VERSION
+
+OPTIONS
+========
+ -d Activate debug mode
+ -h Print help
+
+COMMANDS
+=========
+
+ Host
+ -----
+ host-get-osd hostname List all OSD IDs attached to a particular node.
+ host-get-nodes List all storage nodes.
+ host-osd-usage hostname Show total OSD space usage of a particular node (-d for details).
+ host-all-usage Show total OSD space usage of each nodes (-d for details)
+
+ Placement groups
+ -----------------
+ pg-get-host pgid Find PG storage hosts (first is primary)
+ pg-most-write Find most written PG (nb operations)
+ pg-less-write Find less written PG (nb operations)
+ pg-most-write-kb Find most written PG (data written)
+ pg-less-write-kb Find less written PG (data written)
+ pg-most-read Find most read PG (nb operations)
+ pg-less-read Find less read PG (nb operations)
+ pg-most-read-kb Find most read PG (data read)
+ pg-less-read-kb Find less read PG (data read)
+ pg-empty Find empty PGs (no stored object)
+
+ RBD
+ ----
+ rbd-prefix pool_name image_name Return RBD image prefix
+ rbd-count pool_name image_name Count number of objects in a RBD image
+ rbd-host pool_name image_name Find RBD primary storage hosts
+ rbd-osd pool_name image_name Find RBD primary OSDs
+ rbd-size pool_name image_name Print RBD image real size
+ rbd-all-size pool_name Print all RBD images size (Top first)
+
+ OSD
+ ----
+ osd-most-used Show the most used OSD (capacity)
+ osd-less-used Show the less used OSD (capacity)
+ osd-get-ppg osd_id Show all primaries PGS hosted on a OSD
+ osd-get-pg osd_id Show all PGS hosted on a OSD
+
+ Objects
+ --------
+ object-get-host pool_name object_id Find object storage hosts (first is primary)
+ "
+
+}
+
+#
+# Check dependencies
+#
+function check_requirements()
+{
+
+ # List of command dependencies
+ local bin_dep="ceph rados rbd osdmaptool jq"
+
+ for cmd in $bin_dep; do
+ [ $DEBUG -eq 1 ] && echoinfo "Checking for $cmd..."
+ $cmd --version >/dev/null 2>&1 || { echoerr "$cmd cannot be found... Aborting."; return 1; }
+ done
+
+ CEPH="ceph $CEPH_OPT"
+
+ [ $DEBUG -eq 1 ] && echoinfo "Checking Ceph connectivity & basic permissions..."
+
+ if ! $CEPH -s &> /dev/null; then
+ echoerr "Cannot connect to cluster, please check your username & permissions"
+ echoerr "Command $CEPH -s failed"
+ return 1
+ fi
+
+ JQ="jq -M --raw-output"
+}
+
+#
+# Print the host that hosts a specific PG
+#
+function find_host_from_pg() {
+
+ if [ $# -eq 1 ]; then
+ local PGID=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "PG $PGID has been found at (first is primary) : "
+
+ for osd in $($CEPH pg $PGID query | $JQ -cr .up[]); do
+ echo -n "OSD:osd.$osd | Host:"
+ $CEPH osd find $osd --format json 2> /dev/null | $JQ .crush_location.host
+ done
+}
+
+
+#
+# Print the host that hosts a specific object
+#
+function find_host_from_object() {
+
+ if [ $# -eq 2 ]; then
+ local pool=$1
+ local objid=$2
+ else
+ echoerr "This command requires two arguments"
+ help
+ exit 1
+ fi
+
+ local pgid=$($CEPH osd map $pool $objid --format json 2> /dev/null | $JQ -cr .pgid)
+
+ [ $DEBUG -eq 1 ] && echoinfo $objid found into PG $pgid
+
+ while read host; do
+ echo "PG:$pgid | $host"
+ done < <(find_host_from_pg $pgid)
+}
+
+
+#
+# Print all primary pgs hosted by an OSD
+#
+function find_prim_pg_from_osd() {
+
+ if [ $# -eq 1 ]; then
+ local posd=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Looking for primary PGs beloging to OSD $posd"
+ $CEPH pg dump pgs --format json 2>/dev/null | $JQ --argjson posd $posd '.[] | select(.acting_primary==$posd).pgid'
+}
+
+
+#
+# Print all pgs (primay & secondary) hosted by an OSD
+#
+function find_all_pg_from_osd() {
+
+ if [ $# -eq 1 ]; then
+ local osd=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Looking for all PGs mapped to OSD $osd"
+ $CEPH pg dump pgs --format json 2> /dev/null | $JQ -M --argjson osd $osd '.[] | select(.up[]==$osd).pgid'
+}
+
+
+#
+# Check if a given image exists
+#
+function check_rbd_exists(){
+
+ pool=$1
+ rbd=$2
+
+ if ! rbd info -p $pool $rbd &> /dev/null; then
+ echoerr "Unable to find image $pool/$rbd"
+ exit 1
+ fi
+}
+
+
+#
+# Return RBD prefix from image name
+#
+function get_rbd_prefix() {
+
+ if [ $# -eq 2 ]; then
+ local pool=$1
+ local rbd=$2
+ else
+ echoerr "This command requires two arguments"
+ help
+ exit 1
+ fi
+
+ check_rbd_exists $pool $rbd
+
+ local prefix=$(rbd --image $rbd -p $pool info --format json 2> /dev/null | jq --raw-output .block_name_prefix)
+ if [ -z $prefix ]; then
+ echoerr "Unable to find RBD Prefix for image $pool/$rbd"
+ exit 1
+ else
+ echo $prefix
+ fi
+
+}
+
+
+#
+# Count number of object in a RBD image
+#
+function count_rbd_object() {
+
+ if [ $# -eq 2 ]; then
+ local pool=$1
+ local rbd=$2
+ else
+ echoerr "This command requires two arguments"
+ help
+ exit 1
+ fi
+
+ check_rbd_exists $pool $rbd
+
+ local rbd_prefix=$(get_rbd_prefix $pool $rbd)
+
+ [ $DEBUG -eq 1 ] && echoinfo "RBD image $pool/$rbd has prefix $rbd_prefix; now couning objects..."
+
+ local nb_obj=$(rados -p $pool ls | grep $rbd_prefix | wc -l)
+
+ [ $DEBUG -eq 1 ] && echoinfo "RBD image $pool/$rbd has $nb_obj objects"
+ echo $nb_obj
+}
+
+
+#
+# Find primary storage host for a given RBD image
+#
+function find_prim_host_from_rbd() {
+
+ if [ $# -eq 2 ]; then
+ local pool=$1
+ local rbd=$2
+ else
+ echoerr "This command requires two arguments"
+ help
+ exit 1
+ fi
+
+ check_rbd_exists $pool $rbd
+
+ local osd="null"
+ local osdmap_t=$(mktemp)
+ local osdtree_t=$(mktemp)
+ # Get RBD image prefix
+ local rbd_prefix=$(get_rbd_prefix $pool $rbd)
+# Exit if we received an empty prefix
+ [ -z $rbd_prefix ] && exit 1
+
+# Get pool ID from pool name
+ local pool_id=$(ceph osd lspools -f json | $JQ -M --arg pool $pool '.[]|select(.poolname==$pool).poolnum')
+
+ [ $DEBUG -eq 1 ] && echoinfo "RBD image $pool/$rbd has prefix $rbd_prefix; now finding primary host..."
+
+ [ $DEBUG -eq 1 ] && echoinfo "Dumping OSD map to $osdmap_t"
+ if ! $CEPH osd getmap > $osdmap_t 2> /dev/null; then
+ echoerr "Failed to retrieve OSD map"
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Dumping OSD tree to $osdtree_t"
+
+ if ! $CEPH osd tree --format json > $osdtree_t; then
+ echoerr "Failed to retrieve OSD tree"
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Looking for hosts..."
+
+# For each object in the RBD image
+ for obj in $(rados -p $pool ls | grep $rbd_prefix);
+ do
+# Map object to osd. osdmaptoot does not support json output so using dirty sed.
+ osd=$(osdmaptool --test-map-object $obj --pool $pool_id $osdmap_t 2>/dev/null | sed -r 's/.*\[([[:digit:]]+),.*/\1/' | grep -v osdmaptool)
+# Map osd to host
+ $JQ --argjson osd $osd '.nodes[] | select(.type=="host") | select(.children[] == $osd).name' $osdtree_t
+ done | sort -u
+
+# Cleaning files
+ rm -f $osdtree_t $osdmap_t
+}
+
+
+#
+# Find primary OSDs for a given RBD image
+#
+function find_prim_osd_from_rbd() {
+
+ if [ $# -eq 2 ]; then
+ local pool=$1
+ local rbd=$2
+ else
+ echoerr "This command requires two arguments"
+ help
+ exit 1
+ fi
+
+ check_rbd_exists $pool $rbd
+
+ local osd="null"
+ local osdmap_t=$(mktemp)
+ local osdtree_t=$(mktemp)
+ # Get RBD image prefix
+ local rbd_prefix=$(get_rbd_prefix $pool $rbd)
+
+# Exit if we received an empty prefix
+ [ -z $rbd_prefix ] && exit 1
+
+ [ $DEBUG -eq 1 ] && echoinfo "RBD image $pool/$rbd has prefix $rbd_prefix; now finding primary OSDs..."
+
+ [ $DEBUG -eq 1 ] && echoinfo "Dumping OSD map to $osdmap_t"
+ if ! $CEPH osd getmap > $osdmap_t; then
+ echoerr "Failed to retrieve OSD map"
+ exit 1
+ fi
+
+# For each object in the RBD image
+ for obj in $(rados -p $pool ls | grep $rbd_prefix);
+ do
+# Map object to osd. osdmaptoot does not support json output so using dirty sed.
+ osd=$(osdmaptool --test-map-object $obj $osdmap_t 2>/dev/null | sed -r 's/.*\[([[:digit:]]+),.*/\1/' | grep -v osdmaptool)
+ echo "osd.${osd}"
+ done | sort -u
+
+# Cleaning files
+ rm -f $osdmap_t
+}
+
+
+#
+# Print RBD image real size - Source http://ceph.com/planet/real-size-of-a-ceph-rbd-image/
+#
+
+function print_rbd_real_size {
+
+ if [ $# -eq 2 ]; then
+ local pool=$1
+ local rbd=$2
+ else
+ echoerr "This command requires two arguments"
+ help
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Checking if RBD image exists..."
+
+ check_rbd_exists $pool $rbd
+
+ rbd diff $pool/$rbd | awk '{ SUM += $2 } END { print SUM/1024/1024 " MB" }'
+
+}
+
+
+#
+# Print all RBD image real sizes - Top first
+#
+
+function list_all_rbd_real_size {
+
+ if [ $# -eq 1 ]; then
+ local pool=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Looking for RBD images in pool $pool"
+
+ while read rbd; do
+ [ $DEBUG -eq 1 ] && echoinfo "Inspecting image $rbd"
+ rbd diff $pool/$rbd | awk -v rbd="$rbd" '{ SUM += $2 } END { print SUM/1024/1024 " MB - " rbd }'
+ done < <(rbd -p $pool ls) | sort -rV
+}
+
+
+#
+# Print OSDs belonging to a particular storage host
+#
+
+function list_osd_from_host() {
+
+ if [ $# -eq 1 ]; then
+ local host=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ $CEPH osd tree --format json-pretty 2> /dev/null | $JQ --arg host $host '.nodes[] | select(.type=="host") | select(.name == $host).children[]' | sort -V
+
+}
+
+
+#
+# List all OSD nodes
+#
+
+function list_all_nodes() {
+
+
+ $CEPH osd tree --format json | $JQ -M --raw-output '.nodes[] | select(.type=="host") | .name' | sort -V
+
+}
+
+
+#
+# Print Total OSD usage of a particular storage host
+#
+
+function show_host_osd_usage() {
+
+ if [ $# -eq 1 ]; then
+ local host=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ local pgmap_t=$(mktemp)
+
+ local osd_used_kb=0
+ local total_used_kb=0
+
+ local total_available_kb=0
+ local osd_available_kb=0
+
+ local total_size_kb=0
+ local osd_size_kb=0
+ local nb_osd=0
+
+ [ $DEBUG -eq 1 ] && echoinfo "Dumping PG map..."
+ if ! $CEPH pg dump osds --format json 2>/dev/null > $pgmap_t; then
+ echoerr "Failed to retrieve PG map"
+ exit 1
+ fi
+
+ [ $DEBUG -eq 1 ] && echoinfo "Looking for all OSDs on host $host..."
+
+ for osd in $(list_osd_from_host $host); do
+
+ osd_used_kb=$($JQ --argjson osd $osd '.[] | select(.osd == $osd).kb_used' $pgmap_t)
+ osd_available_kb=$($JQ --argjson osd $osd '.[] | select(.osd == $osd).kb_avail' $pgmap_t)
+ osd_size_kb=$($JQ --argjson osd $osd '.[] | select(.osd == $osd).kb' $pgmap_t)
+
+ [ $DEBUG -eq 1 ] && echoinfo "OSD:$osd | Size:$(echo "scale=1;$osd_size_kb/1024/1024" | bc -l)GB | Used:$(echo "scale=1;$osd_used_kb /1024/1024" | bc -l)GB | Available:$(echo "scale=1;$osd_available_kb/1024/1024" | bc -l)GB"
+
+ let "total_used_kb=total_used_kb+osd_used_kb"
+ let "total_available_kb=total_available_kb+osd_available_kb"
+ let "total_size_kb=total_size_kb+osd_size_kb"
+ let "nb_osd++"
+
+ done
+
+ echo "Host:$host | OSDs:$nb_osd | Total_Size:$(echo "scale=1;$total_size_kb/1024/1024" | bc -l)GB | Total_Used:$(echo "scale=1;$total_used_kb /1024/1024" | bc -l)GB | Total_Available:$(echo "scale=1;$total_available_kb/1024/1024" | bc -l)GB"
+
+ rm -f $pgmap_t
+}
+
+
+#
+# Print Total OSD usage of all nodes
+#
+
+function list_all_nodes_osd_usage() {
+
+
+ for host in $(list_all_nodes); do
+
+ [ $DEBUG -eq 1 ] && echoinfo "Looking at node $host..."
+
+ show_host_osd_usage $host
+ done
+
+}
+
+
+#
+# Find most used (space) OSD
+#
+
+function find_most_used_osd() {
+
+ local osd=$($CEPH pg dump osds --format json 2> /dev/null| $JQ 'max_by(.kb_used) | .osd')
+ local host=$($CEPH osd find $osd 2> /dev/null | $JQ .crush_location.host)
+
+ echo "OSD:osd.${osd} | host:$host"
+}
+
+
+#
+# Find less used (space) OSD
+#
+
+function find_less_used_osd() {
+
+ local osd=$($CEPH pg dump osds --format json 2> /dev/null| $JQ 'min_by(.kb_used) | .osd')
+ local host=$($CEPH osd find $osd 2> /dev/null | $JQ .crush_location.host)
+
+ echo "OSD:osd.${osd} | host:$host"
+}
+
+
+#
+# Query PG stats
+#
+
+function pg_stat_query() {
+
+ if [ $# -eq 1 ]; then
+ local query_type=$1
+ else
+ echoerr "This command requires one argument"
+ help
+ exit 1
+ fi
+
+ local pgmap_t=$(mktemp)
+
+ [ $DEBUG -eq 1 ] && echoinfo "Dumping PG map..."
+ if ! $CEPH pg dump pgs --format json 2>/dev/null > $pgmap_t; then
+ echoerr "Failed to retrieve PG map"
+ exit 1
+ fi
+
+ local pgid=$($JQ --arg query_type $query_type "$query_type" $pgmap_t)
+ [ $DEBUG -eq 1 ] && echoinfo "Found PGID $pgid"
+
+ local osd=$($JQ --arg pgid $pgid '.[] | select(.pgid == $pgid).acting_primary' $pgmap_t)
+ [ $DEBUG -eq 1 ] && echoinfo "Found OSD $osd"
+
+ local host=$($CEPH osd find $osd --format json 2> /dev/null | $JQ .crush_location.host)
+ [ $DEBUG -eq 1 ] && echoinfo "Found host $host"
+
+ echo "PG:$pgid | OSD:osd.$osd | Host:$host"
+
+ rm -f $pgmap_t
+}
+
+
+#
+# Find empty pgs (no object stored)
+#
+
+function find_empty_pg() {
+
+ $CEPH pg dump pgs --format json 2>/dev/null | $JQ '.[] | select(.stat_sum.num_objects == 0).pgid'
+
+}
+
+
+#
+# MAIN
+#
+
+
+# Print help if no argument is given
+if [ $# -eq 0 ]; then
+ help
+ exit 1
+fi
+
+# Activate debug mode if -d is specified as first parameter
+if [ "$1" = "-d" ]; then
+ echoinfo "Debug mode activated"
+ DEBUG=1
+ shift
+else
+ DEBUG=0
+fi
+
+
+# Check if all requirements are met
+check_requirements || exit 1
+
+
+# Call proper function
+case $1 in
+ "-h")
+ help
+ exit 0
+ ;;
+ "host-get-osd")
+ list_osd_from_host $2
+ ;;
+ "host-get-nodes")
+ list_all_nodes
+ ;;
+ "host-osd-usage")
+ show_host_osd_usage $2
+ ;;
+ "host-all-usage")
+ list_all_nodes_osd_usage
+ ;;
+ "pg-get-host")
+ find_host_from_pg $2
+ ;;
+ "pg-most-write")
+ pg_stat_query "max_by(.stat_sum.num_write).pgid"
+ ;;
+ "pg-less-write")
+ pg_stat_query "min_by(.stat_sum.num_write).pgid"
+ ;;
+ "pg-most-write-kb")
+ pg_stat_query "max_by(.stat_sum.num_write_kb).pgid"
+ ;;
+ "pg-less-write-kb")
+ pg_stat_query "min_by(.stat_sum.num_write_kb).pgid"
+ ;;
+ "pg-most-read")
+ pg_stat_query "max_by(.stat_sum.num_read).pgid"
+ ;;
+ "pg-less-read")
+ pg_stat_query "min_by(.stat_sum.num_read).pgid"
+ ;;
+ "pg-most-read-kb")
+ pg_stat_query "max_by(.stat_sum.num_read_kb).pgid"
+ ;;
+ "pg-less-read-kb")
+ pg_stat_query "min_by(.stat_sum.num_read_kb).pgid"
+ ;;
+ "rbd-prefix")
+ get_rbd_prefix $2 $3
+ ;;
+ "rbd-count")
+ count_rbd_object $2 $3
+ ;;
+ "rbd-host")
+ find_prim_host_from_rbd $2 $3
+ ;;
+ "rbd-osd")
+ find_prim_osd_from_rbd $2 $3
+ ;;
+ "rbd-size")
+ print_rbd_real_size $2 $3
+ ;;
+ "rbd-all-size")
+ list_all_rbd_real_size $2
+ ;;
+ "osd-most-used")
+ find_most_used_osd
+ ;;
+ "osd-less-used")
+ find_less_used_osd
+ ;;
+ "osd-get-ppg")
+ find_prim_pg_from_osd $2
+ ;;
+ "osd-get-pg")
+ find_all_pg_from_osd $2
+ ;;
+ "pg-empty")
+ find_empty_pg
+ ;;
+ "object-get-host")
+ find_host_from_object $2 $3
+ ;;
+ *)
+ echoerr "Unknown command : $1"
+ help
+ exit 1
+ ;;
+esac
+
Code Review / stor4nfv.git / blobdiff