SOLR-6313: Improve SolrCloud cloud-dev scripts.

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1616275 13f79535-47bb-0310-9956-ffa450edef68

solr/CHANGES.txt

@@ -330,6 +330,8 @@ Other Changes
 
 * SOLR-5664: Added support for multi-valued field highlighting in /browse UI. (ehatcher)
 
+* SOLR-6313: Improve SolrCloud cloud-dev scripts. (Mark Miller, Vamsee Yarlagadda)
+
 ==================  4.9.0 ==================
 
 Versions of Major Components

solr/cloud-dev/control.sh

@@ -4,7 +4,7 @@ source ./functions.sh
 
 case "$1" in
   start)
-        start $2 $3
+        start $2 $3 "$4"
         ;;
   stop)
         stop $2

solr/cloud-dev/functions.sh

@@ -1,7 +1,7 @@
-JAVA_OPTS="-server -Xms256M -Xmx256M"
+INT_JAVA_OPTS="-server -Xms256M -Xmx256M"
-BASE_PORT=7572
+BASE_PORT=8900
-BASE_STOP_PORT=6572
+BASE_STOP_PORT=9900
-ZK_PORT="9983"
+ZK_PORT="2414"
 
 rebuild() {
 	echo "Rebuilding"
@@ -14,13 +14,8 @@ rebuild() {
 }
 
 setports() {
-	if [ "1" = "$1" ]; then
+  PORT="$(( $BASE_PORT + $1 ))"
-		PORT="8983"
+  STOP_PORT="$(( $BASE_STOP_PORT + $1 ))"
-	    STOP_PORT="7983"
-	else
- 		PORT="$(( $BASE_PORT + $1 ))"
-	        STOP_PORT="$(( $BASE_STOP_PORT + $1 ))"
-	fi
 }
 
 reinstall() {

solr/cloud-dev/solrcloud-start-existing.sh

@@ -1,29 +1,34 @@
 #!/bin/bash
 
 numServers=$1
-baseJettyPort=7572
+
-baseStopPort=6572
+baseJettyPort=8900
+baseStopPort=9900
 
 die () {
     echo >&2 "$@"
     exit 1
 }
 
-[ "$#" -eq 1 ] || die "1 argument required, $# provided, usage: solrcloud-start-exisiting.sh {numServers}"
+[ "$#" -eq 1 ] || die "1 argument required, $# provided, usage: solrcloud-start-exisiting.sh [numServers]"
 
 
 cd ..
 
-cd example1
+cd examplezk
-echo "starting example1"
+stopPort=1313
-java -DzkRun -DSTOP.PORT=7983 -DSTOP.KEY=key -jar start.jar 1>example1.log 2>&1 &
+jettyPort=8900
+exec -a jettyzk java -Xmx512m $JAVA_OPTS -Djetty.port=$jettyPort -DhostPort=$jettyPort -DzkRun -DzkRunOnly=true -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar 1>examplezk.log 2>&1 &
+# TODO: we could also remove the default core
+cd ..
 
+cd example
 
-for (( i=2; i <= $numServers; i++ ))
+for (( i=1; i <= $numServers; i++ ))
 do
   echo "starting example$i"
   cd ../example$i
   stopPort=`expr $baseStopPort + $i`
   jettyPort=`expr $baseJettyPort + $i`
-  java -Xmx1g -Djetty.port=$jettyPort -DzkHost=localhost:9983 -DnumShards=1 -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar 1>example$i.log 2>&1 &
+  exec -a jetty java -Xmx1g $JAVA_OPTS -Djetty.port=$jettyPort -DzkHost=localhost:9900 -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar 1>example$i.log 2>&1 &
 done

solr/cloud-dev/solrcloud-start.sh

@@ -1,19 +1,22 @@
 #!/bin/bash
 
+# To run on hdfs, try something along the lines of:
+# export JAVA_OPTS="-Dsolr.directoryFactory=solr.HdfsDirectoryFactory -Dsolr.lock.type=hdfs -Dsolr.hdfs.home=hdfs://localhost:8020/solr -Dsolr.hdfs.confdir=/etc/hadoop_conf/conf"
+
 numServers=$1
 numShards=$2
 
-baseJettyPort=7572
+baseJettyPort=8900
-baseStopPort=6572
+baseStopPort=9900
 
-zkaddress = localhost:2181/solr
+zkAddress=localhost:9900
 
 die () {
     echo >&2 "$@"
     exit 1
 }
 
-[ "$#" -eq 2 ] || die "2 arguments required, $# provided, usage: solrcloud-start.sh {numServers} {numShards}"
+[ "$#" -eq 2 ] || die "2 arguments required, $# provided, usage: solrcloud-start.sh [numServers] [numShards]"
 
 cd ..
 
@@ -29,6 +32,7 @@ rm -r -f example/solr/zoo_data
 rm -r -f example/solr/collection1/data
 rm -f example/example.log
 
+ant -f ../build.xml clean
 ant example dist
 
 rm -r example/solr-webapp/*
@@ -40,23 +44,28 @@ do
  cp -r -f example example$i
 done
 
+rm -r -f examplezk
+cp -r -f example examplezk
+cp core/src/test-files/solr/solr-no-core.xml examplezk/solr/solr.xml
+rm -r -f examplezk/solr/collection1/core.properties
+cd examplezk
+stopPort=1313
+jettyPort=8900
+exec -a jettyzk java -Xmx512m $JAVA_OPTS -Djetty.port=$jettyPort -DhostPort=$jettyPort -DzkRun -DzkRunOnly=true -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar 1>examplezk.log 2>&1 &
+cd ..
 
-java -classpath "example1/solr-webapp/webapp/WEB-INF/lib/*:example/lib/ext/*" org.apache.solr.cloud.ZkCLI -cmd bootstrap -zkhost 127.0.0.1:9983 -solrhome example1/solr -runzk 8983
+# upload config files
+java -classpath "example1/solr-webapp/webapp/WEB-INF/lib/*:example/lib/ext/*" org.apache.solr.cloud.ZkCLI -cmd bootstrap -zkhost $zkAddress -solrhome example1/solr
   
-echo "starting example1"
+cd example
 
-cd example1
+for (( i=1; i <= $numServers; i++ ))
-java -Xmx1g -DzkRun -DnumShards=$numShards -DSTOP.PORT=7983 -DSTOP.KEY=key -jar start.jar 1>example1.log 2>&1 &
-
-
-
-for (( i=2; i <= $numServers; i++ ))
 do
   echo "starting example$i"
   cd ../example$i
   stopPort=`expr $baseStopPort + $i`
   jettyPort=`expr $baseJettyPort + $i`
-  java -Xmx1g -Djetty.port=$jettyPort -DzkHost=localhost:9983 -DnumShards=1 -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar 1>example$i.log 2>&1 &
+  exec -a jetty java -Xmx1g $JAVA_OPTS -DnumShards=$numShards -Djetty.port=$jettyPort -DzkHost=$zkAddress -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar 1>example$i.log 2>&1 &
 done
 
 

solr/cloud-dev/stop.sh

@@ -1,7 +1,8 @@
 #!/bin/bash
 
 numServers=$1
-baseStopPort=6572
+baseJettyPort=8900
+baseStopPort=9900
 
 die () {
     echo >&2 "$@"
@@ -12,13 +13,52 @@ die () {
 
 cd ../example
 
-java -DSTOP.PORT=7983 -DSTOP.KEY=key -jar start.jar --stop
+for (( i=1; i <= $numServers; i++ ))
-
-
-for (( i=2; i <= $numServers; i++ ))
 do
-  echo "stopping example$i"
-  cd ../example$i
   stopPort=`expr $baseStopPort + $i`
+  echo "stopping example$i, stop port is $stopPort"
+  cd ../example$i
   java -DSTOP.PORT=$stopPort -DSTOP.KEY=key -jar start.jar --stop
 done
+
+
+mkdir ../example-lastlogs
+
+for (( i=1; i <= $numServers; i++ ))
+do
+   cd ../example$i
+
+  jettyPort=`expr $baseJettyPort + $i`
+  echo "Make sure jetty stops and wait for it: $jettyPort"
+
+  pid=`lsof -i:$jettyPort -sTCP:LISTEN -t`
+  echo "pid:$pid"
+  #kill $pid
+  #wait $pid
+  if [ ! -z "$pid" ]
+  then
+    while [ -e /proc/$pid ]; do sleep 1; done
+  fi
+  
+  # save the last shutdown logs
+  echo "copy example$i.log to lastlogs"
+  cp -r -f example$i.log ../example-lastlogs/example-last$i.log
+done
+
+# stop zk runner
+java -DSTOP.PORT=1313 -DSTOP.KEY=key -jar start.jar --stop
+
+echo "wait for port to be available: $baseJettyPort"
+
+pid=`lsof -i:$baseJettyPort -sTCP:LISTEN -t`
+echo "pid:$pid"
+#kill $pid
+#wait $pid
+if [ ! -z "$pid" ]
+then
+  while [ -e /proc/$pid ]; do sleep 0.1; done
+fi
+nc -w 30 127.0.0.1 $baseJettyPort
+
+sleep 5
+ 

solr/core/src/java/org/apache/solr/cloud/ZkController.java

@@ -178,6 +178,12 @@ public final class ZkController {
   // keeps track of replicas that have been asked to recover by leaders running on this node
   private Map<String,String> replicasInLeaderInitiatedRecovery = new HashMap<String,String>();  
   
+  // This is an expert and unsupported development mode that does not create
+  // an Overseer or register a /live node. This let's you monitor the cluster
+  // and interact with zookeeper via the Solr admin UI on a node outside the cluster,
+  // and so one that will not be killed or stopped when testing. See developer cloud-scripts.
+  private boolean zkRunOnly = Boolean.getBoolean("zkRunOnly"); // expert
+  
   public ZkController(final CoreContainer cc, String zkServerAddress, int zkClientTimeout, int zkClientConnectTimeout, String localHost, String locaHostPort,
         String localHostContext, int leaderVoteWait, int leaderConflictResolveWait, boolean genericCoreNodeNames, final CurrentCoreDescriptorProvider registerOnReconnect) 
       throws InterruptedException, TimeoutException, IOException
@@ -228,16 +234,19 @@ public final class ZkController {
               
               registerAllCoresAsDown(registerOnReconnect, false);
               
-              ElectionContext context = new OverseerElectionContext(zkClient,
+              if (!zkRunOnly) {
-                  overseer, getNodeName());
+                ElectionContext context = new OverseerElectionContext(zkClient,
+                    overseer, getNodeName());
                 
-              ElectionContext prevContext = overseerElector.getContext();
+                ElectionContext prevContext = overseerElector.getContext();
-              if (prevContext != null) {
+                if (prevContext != null) {
-                prevContext.cancelElection();
+                  prevContext.cancelElection();
+                }
+                
+                overseerElector.setup(context);
+                overseerElector.joinElection(context, true);
               }
               
-              overseerElector.setup(context);
-              overseerElector.joinElection(context, true);
               zkStateReader.createClusterStateWatchersAndUpdate();
               
               // we have to register as live first to pick up docs in the buffer
@@ -686,6 +695,9 @@ public final class ZkController {
 
   private void createEphemeralLiveNode() throws KeeperException,
       InterruptedException {
+    if (zkRunOnly) {
+      return;
+    }
     String nodeName = getNodeName();
     String nodePath = ZkStateReader.LIVE_NODES_ZKNODE + "/" + nodeName;
     log.info("Register node as live in ZooKeeper:" + nodePath);