IMPALA-9165: Add back hard kill to kill-hbase.sh
The fix for IMPALA-9150 changed kill-hbase.sh to use HBase's
stop-hbase.sh script. Around this time, the GVO timeout issues
started. GVO can reuse machines, so we don't know what state
they may be in. If something failed to kill HBase processes,
the next job would need to be able to kill them even without
access to the last run's files / logs.
This restores the original kill logic to kill-hbase.sh, after
trying a graceful shutdown using HBase's stop-hbase.sh script.
The original kill logic doesn't rely on anything from the
filesystem to know about the existence of processes, so it
would handle machine reuse.
This also changes our Jenkins test scripts to shut down the
minicluster at the end.
Testing:
- Started with a running minicluster, ran bin/clean.sh,
then ran testdata/bin/kill-all.sh and verified that the
java processes were gone
Change-Id: Ie2f0b342bcd1d8abea8ef923adbb54a14518a7a6
Reviewed-on: http://gerrit.cloudera.org:8080/14789
Reviewed-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
diff --git a/bin/jenkins/all-tests.sh b/bin/jenkins/all-tests.sh
index b531804..21c3acc 100644
--- a/bin/jenkins/all-tests.sh
+++ b/bin/jenkins/all-tests.sh
@@ -67,5 +67,8 @@
RET_CODE=1
fi
+# Shutdown minicluster at the end
+testdata/bin/kill-all.sh
+
bin/jenkins/finalize.sh
exit $RET_CODE
diff --git a/bin/jenkins/dockerized-impala-run-tests.sh b/bin/jenkins/dockerized-impala-run-tests.sh
index e097fcc..45f7e86 100755
--- a/bin/jenkins/dockerized-impala-run-tests.sh
+++ b/bin/jenkins/dockerized-impala-run-tests.sh
@@ -92,4 +92,12 @@
# to make them start up dockerised clusters and will probably make more assumptions
# about the cluster being tested.
export CLUSTER_TEST=false
-./bin/run-all-tests.sh
+RET_CODE=0
+if ! ./bin/run-all-tests.sh; then
+ RET_CODE=1
+fi
+
+# Shutdown minicluster at the end
+./testdata/bin/kill-all.sh
+
+exit $RET_CODE
diff --git a/testdata/bin/kill-hbase.sh b/testdata/bin/kill-hbase.sh
index e404aef..47298c8 100755
--- a/testdata/bin/kill-hbase.sh
+++ b/testdata/bin/kill-hbase.sh
@@ -23,10 +23,14 @@
DIR=$(dirname "$0")
echo Stopping Hbase
-# Use the stop-hbase.sh script provided by HBase. This does a more graceful shutdown
-# that using our kill-java-service.sh script.
+# First, use the stop-hbase.sh script provided by HBase. This does a more graceful
+# shutdown than using our kill-java-service.sh script.
${HBASE_HOME}/bin/stop-hbase.sh
+# Second, also do a harder shutdown in case there is anything that is not covered
+# by the stop-hbase.sh script. Kill region server first, then hmaster, and zookeeper.
+"$DIR"/kill-java-service.sh -c HRegionServer -c HMaster -c HQuorumPeer -s 2
+
# Clear up data so that zookeeper/hbase won't do recovery when it starts.
# TODO: is this still needed when using bin/stop-hbase.sh?
rm -rf /tmp/hbase-*
