[jira] [Work logged] (HIVE-22193) Graceful Shutdown HiveServer2

[ASF GitHub Bot (Jira)]

Title: Message Title

ASF GitHub Bot added a worklog on HIVE-22193   dengzhhu653 commented on code in PR #3386: URL: https://github.com/apache/hive/pull/3386#discussion_r908002850 ########## bin/ext/hiveserver2.sh: ########## @@ -16,24 +16,83 @@ THISSERVICE=hiveserver2 export SERVICE_LIST="${SERVICE_LIST}${THISSERVICE} " +if [ "$HIVESERVER2_PID_DIR" = "" ]; then + HIVESERVER2_PID_DIR=$HIVE_CONF_DIR +fi + +HIVESERVER2_PID=$HIVESERVER2_PID_DIR/hiveserver2.pid + +before_start() { + #ckeck if the process is not running + mkdir -p "$HIVESERVER2_PID_DIR" + if [ -f $HIVESERVER2_PID ]; then + if kill -0 $(cat $HIVESERVER2_PID) >/dev/null 2>&1; then + echo "HiveServer2 running as process $(cat $HIVESERVER2_PID). Stop it first." + exit 1 + fi + fi +} + hiveserver2() { >&2 echo "$(timestamp): Starting HiveServer2" CLASS=org.apache.hive.service.server.HiveServer2 if $cygwin; then HIVE_LIB=`cygpath -w "$HIVE_LIB"` + if [ "$1" = "--graceful_stop" ]; then + pid=$2 + if [ "$pid" = "" -a -f $HIVESERVER2_PID ]; then + pid=$(cat $HIVESERVER2_PID) + fi + if [ "$pid" != "" ]; then + timeout=$3 + killAndWait $pid ${timeout:-1800} + fi + else + before_start + echo >&2 "$(timestamp): Starting HiveServer2" + CLASS=org.apache.hive.service.server.HiveServer2 + if $cygwin; then + HIVE_LIB=$(cygpath -w "$HIVE_LIB") + fi + JAR=${HIVE_LIB}/hive-service-[0-9].*.jar + + export HADOOP_CLIENT_OPTS=" -Dproc_hiveserver2 $HADOOP_CLIENT_OPTS " + export HADOOP_OPTS="$HIVESERVER2_HADOOP_OPTS $HADOOP_OPTS" + hiveserver2_pid="$$" Review Comment: > so this would be the PID of the current shell script that starts HS2 right? At this point, we havent launched the HS2 process yet yeah, that's right. the `exec` command will inherit the process ID of the parent, we also have used in [hive](https://github.com/apache/hive/blob/master/bin/hive#L383) to get the pid of the service process. > Dont we need the PID of HS2 process? if so, should HS2 write this file? also, what happens when there are multiple HS2 instances? yes, the pid is used later for killing the process, so we should persist it into a file where locates at $HIVE_CONF_DIR/hiveserver2.pid by default, usually we have a single corresponding `HIVE_CONF_DIR` for each HS2 instance, and this can be changed via env `HIVESERVER2_PID_DIR` when starting the HS2. Also add a [check](https://github.com/apache/hive/blob/23452342c84cef625b1b89f5e3e9e16bbe9103ad/bin/ext/hiveserver2.sh#L25-L34) here before we start HS2, this avoid mutiple HS2 instances writing its pid to the same file. Worklog entry: 10m logged by ASF GitHub Bot Change By: ASF GitHub Bot Time Spent: 40m 50m Add Comment   This message was sent by Atlassian Jira (v8.20.10#820010-sha1:ace47f9)