[hotfix][tests] Shut down HDFS and YARN cleanly.

Use provided scripts to shut down HDFS and YARN instead of grepkill! function.
Using grepkill! one can run situations where flink and hadoop processes are
killed concurrently, which results in an exception.

flink-jepsen/src/jepsen/flink/hadoop.clj

@@ -115,18 +115,37 @@
     (info "Start DataNode")
     (c/exec (str install-dir "/sbin/hadoop-daemon.sh") :--config hadoop-conf-dir :--script :hdfs :start :datanode)))
 
+(defn stop-data-node!
+  []
+  (c/exec (str install-dir "/sbin/hadoop-daemon.sh") :--config hadoop-conf-dir :--script :hdfs :stop :datanode))
+
 (defn start-resource-manager!
   [test node]
   (when (= node (resource-manager (:nodes test)))
     (info "Start ResourceManager")
     (c/exec (str install-dir "/sbin/yarn-daemon.sh") :--config hadoop-conf-dir :start :resourcemanager)))
 
+(defn stop-resource-manager!
+  []
+  (c/exec (str install-dir "/sbin/yarn-daemon.sh") :--config hadoop-conf-dir :stop :resourcemanager))
+
 (defn start-node-manager!
   [test node]
   (when (some #{node} (data-nodes (:nodes test)))
     (info "Start NodeManager")
     (c/exec (str install-dir "/sbin/yarn-daemon.sh") :--config hadoop-conf-dir :start :nodemanager)))
 
+(defn stop-node-manager!
+  []
+  (c/exec (str install-dir "/sbin/yarn-daemon.sh") :--config hadoop-conf-dir :stop :nodemanager))
+
+(defn stop-all!
+  []
+  (stop-name-node!)
+  (stop-data-node!)
+  (stop-resource-manager!)
+  (stop-node-manager!))
+
 (defn db
   [url]
   (reify db/DB
@@ -143,7 +162,7 @@
     (teardown! [_ _ _]
       (info "Teardown Hadoop")
       (c/su
-        (cu/grepkill! "hadoop")
+        (stop-all!)
         (c/exec :rm :-rf install-dir)))
 
     db/LogFiles