@@ -52,6 +52,7 @@ $(cyan Commands):
5252 $( green stop)
5353 $( green restart)
5454 $( green kill)
55+ $( green prune)
5556
5657$( cyan Examples)
5758 $( purple ' accumulo-cluster start' ) $( blue ' # start all servers' )
@@ -62,6 +63,8 @@ $(cyan Examples):
6263 $( purple ' accumulo-cluster start --sservers=group1' ) $( blue ' # start all group1 sservers' )
6364 $( purple ' accumulo-cluster start --sservers="group1 group2"' ) $( blue ' # start all group1 and group2 sservers' )
6465 $( purple ' accumulo-cluster start --local --manager --tservers' ) $( blue ' # Start the local manager and local tservers' )
66+ $( purple ' accumulo-cluster prune --compactors' ) $( blue ' # prune all extra compactors across all groups' )
67+ $( purple ' accumulo-cluster prune --compactors="group1"' ) $( blue ' # prune extra compactors running in group1' )
6568
6669EOF
6770}
@@ -287,8 +290,14 @@ function parse_config() {
287290 exit 1
288291 fi
289292
290- trap ' rm -f "$CONFIG_FILE"'
291- CONFIG_FILE=$( mktemp --tmpdir " ClusterConfigParser-XXXXXXXX.out" ) || exit 1
293+ AC_TMP_DIR=$( mktemp -t -d " accumulo-cluster-XXXXXXXX" ) || exit 1
294+ if isDebug; then
295+ echo " $( blue DEBUG) $AC_TMP_DIR "
296+ else
297+ trap ' rm -rf -- "$AC_TMP_DIR"'
298+ fi
299+
300+ CONFIG_FILE=" $AC_TMP_DIR /ClusterConfigParser.out"
292301 " $accumulo_cmd " " $conf /cluster.yaml" " $CONFIG_FILE " || parse_fail
293302 # shellcheck source=/dev/null
294303 . " $CONFIG_FILE "
@@ -590,9 +599,183 @@ function control_services() {
590599
591600 if [[ $ARG_LOCAL == 0 && $ARG_ALL == 1 && ($operation == " stop" || $operation == " kill" ; then
592601 debug " Cleaning all server entries in ZooKeeper"
593- debugOrRun " $accumulo_cmd "
602+ debugOrRun " $accumulo_cmd "
603+ fi
604+
605+ }
606+
607+ function prune_group() {
608+ local service_type=$1
609+ local group=$2
610+ local expectedCount=$3
611+ declare -a hosts
612+ read -r -a hosts <<< " $4"
613+
614+ if isDebug; then
615+ echo " $( blue DEBUG) $service_type group:$group expected:$expectedCount "
616+ fi
617+
618+ if [ -z ${AC_TMP_DIR+x} ]; then
619+ echo " $( red ERROR) "
620+ exit 1
621+ fi
622+ local exclude_file=" $AC_TMP_DIR /accumulo-zoozap-exclude-$service_type -$group .txt"
623+ touch " $exclude_file "
624+
625+ # Determine the host:ports known by the accumulo cluster script, these should be kept
626+ for host in " ${hosts[@]} " ; do
627+ " ${SSH[@]} " " $host " " '$bin /accumulo-service $service_type list'" | grep -E " ^[a-zA-Z0-9]+_${group} _[0-9]+" | head -n " $expectedCount " | awk ' {print $3}' | tr ' ,' ' \n' | awk ' {print "' " $host " ' :" $1}' >> " $exclude_file "
628+ done
629+
630+ local lockTypeOpt
631+ case $service_type in
632+ manager)
633+ lockTypeOpt=" -manager"
634+ ;;
635+ compaction-coordinator)
636+ lockTypeOpt=" -compaction-coordinators"
637+ ;;
638+ compactor)
639+ lockTypeOpt=" -compactors"
640+ ;;
641+ tserver)
642+ lockTypeOpt=" -tservers"
643+ ;;
644+ sserver)
645+ lockTypeOpt=" -sservers"
646+ ;;
647+ gc)
648+ lockTypeOpt=" --gc"
649+ ;;
650+ monitor)
651+ lockTypeOpt=" --monitor"
652+ ;;
653+ * )
654+ echo " Prune does not support $service_type "
655+ exit 1
656+ ;;
657+ esac
658+
659+ if isDebug; then
660+ " $accumulo_cmd " " $lockTypeOpt " " $group " " $exclude_file "
661+ else
662+ " $accumulo_cmd " " $lockTypeOpt " " $group " " $exclude_file "
663+ fi
664+ }
665+
666+ # Kills extra server processes that are not needed according to the
667+ # cluster.yaml file. Conceptually this code is trying to reconcile the
668+ # following three sets of servers.
669+ #
670+ # 1. The notional goal set of servers specified by cluster.yaml
671+ # 2. The set of servers processes seen in zookeeper
672+ # 3. The set of server processes known to the accumulo-cluster script. This
673+ # is derived from pid files on hosts in set 1.
674+ #
675+ # This function attempts to find extra servers in set 2 that are not specified
676+ # by set 1. When it does find extra servers it removes their zookeeper locks
677+ # avoiding removing locks of servers in set 3. The following are different
678+ # situations the code will see and handle.
679+ #
680+ # * When a host is not cluster.yaml but has some processes listed in
681+ # zookeeper. For this case all of the process with that host can be killed.
682+ # * When a resource group is not in cluster.yaml but has some processes listed
683+ # in zookeeper. For this case all of the processes with that resource group
684+ # can be killed.
685+ # * When a host is in cluster.yaml with a target of 3 processes but has 6
686+ # processes listed in zookeeper. For this case want to kill 3 processes that
687+ # do not have pid files on the host.
688+ #
689+ function prune() {
690+ if [[ $ARG_LOCAL == 1 ]]; then
691+ # Currently the code is structured to remove all extra servers in a single resource group. Finer granularity is not supported.
692+ echo " $( red ERROR) "
693+ exit 1
694+ fi
695+
696+ if ! jq -h >& /dev/null; then
697+ echo " $( red ERROR:) $( green jq) "
698+ exit 1
699+ fi
700+
701+ if [[ -z ${AC_TMP_DIR+x} ]]; then
702+ echo " AC_TMP_DIR is not set"
703+ exit 1
704+ fi
705+ local service_json=" $AC_TMP_DIR /accumulo-service.json"
706+ " $accumulo_cmd " > " $service_json " 2> /dev/null || exit 1
707+
708+ local var_name
709+ local hosts
710+ declare -a groups
711+
712+ local manager
713+ if [[ $ARG_ALL == 1 || $ARG_MANAGER == 1 ]]; then
714+ prune_group " manager" " default" " 1" " $MANAGER_HOSTS "
715+ fi
716+
717+ if [[ $ARG_ALL == 1 || $ARG_GC == 1 ]]; then
718+ prune_group " gc" " default" " 1" " $GC_HOSTS "
594719 fi
595720
721+ if [[ $ARG_ALL == 1 || $ARG_MONITOR == 1 ]]; then
722+ prune_group " monitor" " default" " 1" " $MONITOR_HOSTS "
723+ fi
724+
725+ if [[ $ARG_ALL == 1 || $ARG_TSERVER == 1 ]]; then
726+ groups=()
727+ if [[ -n $ARG_TSERVER_GROUP ]]; then
728+ read -r -a groups <<< " $ARG_TSERVER_GROUP"
729+ else
730+ # find all groups known in zookeeper, this will allow pruning entire groups that do not even exist in cluster.yaml
731+ readarray -t groups < <( jq -r " .summaries.T_SERVER.resourceGroups | .[] " " $service_json " )
732+ fi
733+
734+ for group in " ${groups[@]} " ; do
735+ var_name=" TSERVERS_PER_HOST_$group "
736+ local expected=${! var_name:- 0}
737+
738+ hosts=" TSERVER_HOSTS_$group "
739+ prune_group " tserver" " $group " " $expected " " ${! hosts} "
740+ done
741+ fi
742+
743+ if [[ $ARG_ALL == 1 || $ARG_SSERVER == 1 ]]; then
744+ groups=()
745+ if [[ -n $ARG_SSERVER_GROUP ]]; then
746+ read -r -a groups <<< " $ARG_SSERVER_GROUP"
747+ else
748+ # find all groups known in zookeeper, this will allow pruning entire groups that do not even exist in cluster.yaml
749+ readarray -t groups < <( jq -r " .summaries.S_SERVER.resourceGroups | .[] " " $service_json " )
750+ fi
751+
752+ for group in " ${groups[@]} " ; do
753+ var_name=" SSERVERS_PER_HOST_$group "
754+ local expected=${! var_name:- 0}
755+
756+ hosts=" SSERVER_HOSTS_$group "
757+ prune_group " sserver" " $group " " $expected " " ${! hosts} "
758+ done
759+
760+ fi
761+
762+ if [[ $ARG_ALL == 1 || $ARG_COMPACTOR == 1 ]]; then
763+ groups=()
764+ if [[ -n $ARG_COMPACTOR_GROUP ]]; then
765+ read -r -a groups <<< " $ARG_COMPACTOR_GROUP"
766+ else
767+ # find all groups known in zookeeper, this will allow pruning entire groups that do not even exist in cluster.yaml
768+ readarray -t groups < <( jq -r " .summaries.COMPACTOR.resourceGroups | .[] " " $service_json " )
769+ fi
770+
771+ for group in " ${groups[@]} " ; do
772+ var_name=" COMPACTORS_PER_HOST_$group "
773+ local expected=${! var_name:- 0}
774+
775+ hosts=" COMPACTOR_HOSTS_$group "
776+ prune_group " compactor" " $group " " $expected " " ${! hosts} "
777+ done
778+ fi
596779}
597780
598781function main() {
674857 parse_config
675858 control_services kill
676859 ;;
860+ prune)
861+ parse_config
862+ prune
863+ ;;
677864 * )
678865 invalid_args " '$ARG_CMD ' is an invalid <command>"
679866 ;;
0 commit comments