Updated packages for MOSK 25.1 release, ceph collect scipt
Related-PROD: PROD-37028
Change-Id: I863ffc55a0a8ead0562cc3c104a5f209ac320fb2
diff --git a/scripts/ceph_collect.sh b/scripts/ceph_collect.sh
index fe2ef6c..1285aa9 100644
--- a/scripts/ceph_collect.sh
+++ b/scripts/ceph_collect.sh
@@ -1,9 +1,58 @@
#!/bin/bash
+
+# Collect relevant data from a Ceph cluster for troubleshooting and assessment
+# (C) Christian Huebner chuebner@mirantis.com 2015
+# run with ./ceph_collect.sh <customername> <clustername>
+
echo "Collecting Ceph cluster data."
-if [ "$#" -lt 2 ]; then echo "Usage: ./ceph_collect.sh <CUSTOMER> <CLUSTERNAME>"; exit; fi
-export CUSTOMER=$1
-export CLUSTERNAME=$2
+help () {
+ echo "Data collector for Ceph analytics"
+ echo "Usage: ceph_collect.sh [-b] [-m] <customername> <clustername>"
+ echo "-m only works with Nautilus and up"
+}
+
+POSITIONAL=()
+BENCH="true"
+VOLUMES="false"
+OSDS=(`ceph osd ls`)
+
+if [[ $# -eq 0 ]]; then
+ help
+ exit
+fi
+
+while [[ $# -gt 0 ]]; do
+ key="$1"
+
+ case $key in
+ -h|--help)
+ help
+ exit
+ ;;
+ -vol|--volumes)
+ VOLUMES="true"
+ shift
+ ;;
+ -nb|--nobench)
+ BENCH="false"
+ shift # past argument
+ ;;
+ -m|--healthmetrics)
+ HEALTHMETRICS="true"
+ shift # past argument
+ ;;
+ *) # unknown option
+ POSITIONAL+=("$1") # save it in an array for later
+ shift # past argument
+ ;;
+ esac
+done
+
+if [ ${#POSITIONAL[*]} -lt 2 ]; then echo "Usage: ./ceph_collect.sh <CUSTOMER> <CLUSTERNAME>"; exit; fi
+# if [ "$#" -lt 2 ]; then echo "Usage: ./ceph_collect.sh <CUSTOMER> <CLUSTERNAME>"; exit; fi
+export CUSTOMER=${POSITIONAL[0]}
+export CLUSTERNAME=${POSITIONAL[1]}
if ! which ceph >/dev/null; then echo "ERROR: This script must be run on a ceph monitor or admin node"; exit; fi
@@ -25,7 +74,7 @@
echo "Collecting cluster status"
ceph -s -f json -o ceph_s.json
echo "Collecting health detail"
-ceph -f json health detail -o ceph_healt_detail.json
+ceph -f json health detail -o ceph_health_detail.json
echo "Collecting monmap"
ceph mon dump -f json -o monmap.json
echo "Collecting ceph df"
@@ -42,13 +91,81 @@
ceph auth list -f json |sed 's/AQ[^=]*==/KEY/g' > ceph_auth_ls.json
echo "Collecting ceph pg dump"
ceph pg dump -f json -o ceph_pg_dump.json
-echo "Collecting health metrics"
-mkdir ceph-health
-IFS=$'\n'; for device in `ceph device ls|grep -v DEVICE`; do osd=$(echo $device|awk '{print $3}'); dev=$(echo $device|awk '{print $1}'); ceph device get-health-metrics $dev >ceph-health/$osd-$dev.json ; done
-echo "Collecting ceph osd perf"
-for i in {0..9}; do echo $i; ceph osd perf -f json -o ceph_osd_perf_$i.json; sleep 4; done
+echo "Collecting ceph pg autoscale"
+ceph osd pool autoscale-status -f json -o ceph_pg_autoscale_status.json
echo "Collecting ceph running configuration"
ceph config dump -f json >ceph_config_dump.json
+echo "Collecting ceph erasure code profiles"
+ceph -f json osd erasure-code-profile ls >ceph_osd_erasure-code-profiles.json
+
+
+echo "Collecting erasure code profiles"
+ceph -f json osd erasure-code-profile ls >ceph_osd_erasure-code-profiles.json
+
+echo "Collecting rbd ls -l"
+rbd ls -l | sed '$ s/.$/}/' >rbd_ls_l.json
+
+echo "Collecting block DB/WAL stats"
+for i in ${OSDS[@]}; do
+ echo \"osd.$i\"\: `ceph -f json tell osd.$i bluefs stats` >>ceph_db_wal_stats.json;
+done
+
+for prof in `ceph osd erasure-code-profile ls`; do \
+ ceph -f json -o ceph_osd_erasure-code-profile_$prof.json osd erasure-code-profile get $prof; \
+done
+
+if [[ $VOLUMES = "true" ]]; then
+ echo "Collecting ceph volumes and CoW clones per rbd pool"
+ for pool in `ceph osd pool ls detail |grep rbd |awk '{print $3}'|sed s/\'//g`; do \
+ echo '{' \
+ '"name" : ' $pool ',' \
+ '"volumes" : ' `rbd ls -l $pool | grep -v '@'| wc -l |grep -v ' 0$'` ',' \
+ '"snapshots" : ' `rbd ls -l $pool | awk '{print $1}' |grep '@' | wc -l` ',' \
+ '"clones" : ' `rbd ls -l $pool | awk '{print $4}' |grep '@' | wc -l` ',' \
+ '}' ; \
+ done >> volumes_per_pool.json
+else
+ echo "Volume collection disabled."
+fi
+
+echo "Collecting ceph osd perf"
+for i in {0..9}; do echo $i; ceph osd perf -f json -o ceph_osd_perf_$i.json; sleep 4; done
+
+if $(ceph device ls &>/dev/null); then
+ echo "Collecting device health information"
+ ceph device ls|grep -v DEVICE|awk '{print $1}'|xargs --max-args=1 ceph device get-health-metrics > ceph_device_get_health_metrics.json
+else
+ echo "Device health check not supported"
+fi
+
+if [[ $HEALTHMETRICS = "true" ]]; then
+ echo "Collecting Ceph Health Metrics (-m option)"
+ IFS=$'\n'
+ for device in `ceph device ls|grep -v DEVICE`; do
+ osd=$(echo $device|awk '{print $3}');
+ dev=$(echo $device|awk '{print $1}');
+ ceph device get-health-metrics $dev >ceph_health_$osd.json ;
+ done;
+fi
+
+if [[ $BENCH = "true" ]]; then
+ echo "Collecting Ceph Benchmark"
+ echo "{ " > ceph_tell_bench.json
+ for i in ${OSDS[@]}; do
+ if [ $i -ne ${OSDS[-1]} ]; then
+ echo \"osd.$i\"\: `ceph tell osd.$i bench -f json 12000000 4096` ',' >>ceph_tell_bench.json;
+ else
+ echo \"osd.$i\"\: `ceph tell osd.$i bench -f json 12000000 4096` >>ceph_tell_bench.json;
+ fi
+ done;
+ echo "}" >> ceph_tell_bench.json
+fi
+
+# Grab Ceph logs
+echo "Collecting Ceph Logs"
+ceph log last 10000 cluster >ceph_cluster_log.json || true
+ceph log last 10000 audit >ceph_audit_log.json || true
tar czf "../"$ARCHNAME *
+cd ..