SOLR-4814: If a SolrCore cannot be created it should remove any information it published about itself from ZooKeeper.

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1485831 13f79535-47bb-0310-9956-ffa450edef68

solr/CHANGES.txt

@@ -105,6 +105,9 @@ Bug Fixes
 * SOLR-4842: Fix facet.field local params from affecting other facet.field's.
   (ehatcher, hossman)
 
+* SOLR-4814: If a SolrCore cannot be created it should remove any information it 
+  published about itself from ZooKeeper. (Mark Miller)
+
 Other Changes
 ----------------------
 

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

@@ -537,6 +537,15 @@ public class Overseer {
         DocCollection coll = newCollections.get(collection);
         if (coll == null) {
           // TODO: log/error that we didn't find it?
+          // just in case, remove the zk collection node
+          try {
+            zkClient.clean("/collections/" + collection);
+          } catch (InterruptedException e) {
+            SolrException.log(log, "Cleaning up collection in zk was interrupted:" + collection, e);
+            Thread.currentThread().interrupt();
+          } catch (KeeperException e) {
+            SolrException.log(log, "Problem cleaning up collection in zk:" + collection, e);
+          }
           return clusterState;
         }
 

solr/core/src/java/org/apache/solr/core/CoreContainer.java

@@ -426,6 +426,16 @@ public class CoreContainer
                   c = create(p);
                   registerCore(p.isTransient(), name, c, false);
                 } catch (Throwable t) {
+                  if (isZooKeeperAware()) {
+                    try {
+                      zkSys.zkController.unregister(name, p);
+                    } catch (InterruptedException e) {
+                      Thread.currentThread().interrupt();
+                      SolrException.log(log, null, e);
+                    } catch (KeeperException e) {
+                      SolrException.log(log, null, e);
+                    }
+                  }
                   SolrException.log(log, null, t);
                   if (c != null) {
                     c.close();
@@ -988,6 +998,19 @@ public class CoreContainer
         core.open();
       }
     } catch(Exception ex){
+      // remains to be seen how transient cores and such
+      // will work in SolrCloud mode, but just to be future 
+      // proof...
+      if (isZooKeeperAware()) {
+        try {
+          getZkController().unregister(name, desc);
+        } catch (InterruptedException e) {
+          Thread.currentThread().interrupt();
+          SolrException.log(log, null, e);
+        } catch (KeeperException e) {
+          SolrException.log(log, null, e);
+        }
+      }
       throw recordAndThrow(name, "Unable to create core: " + name, ex);
     } finally {
       solrCores.removeFromPendingOps(name);

solr/core/src/java/org/apache/solr/handler/admin/CoreAdminHandler.java

@@ -414,6 +414,7 @@ public class CoreAdminHandler extends RequestHandlerBase {
       throw new SolrException(SolrException.ErrorCode.BAD_REQUEST,
                               "Core name is mandatory to CREATE a SolrCore");
     }
+    CoreDescriptor dcore = null;
     try {
       
       if (coreContainer.getAllCoreNames().contains(name)) {
@@ -428,7 +429,7 @@ public class CoreAdminHandler extends RequestHandlerBase {
         instanceDir = name; // bare name is already relative to solr home
       }
 
-      CoreDescriptor dcore = new CoreDescriptor(coreContainer, name, instanceDir);
+      dcore = new CoreDescriptor(coreContainer, name, instanceDir);
 
       //  fillup optional parameters
       String opts = params.get(CoreAdminParams.CONFIG);
@@ -511,6 +512,16 @@ public class CoreAdminHandler extends RequestHandlerBase {
       rsp.add("core", core.getName());
       return coreContainer.isPersistent();
     } catch (Exception ex) {
+      if (coreContainer.isZooKeeperAware() && dcore != null) {
+        try {
+          coreContainer.getZkController().unregister(name, dcore);
+        } catch (InterruptedException e) {
+          Thread.currentThread().interrupt();
+          SolrException.log(log, null, e);
+        } catch (KeeperException e) {
+          SolrException.log(log, null, e);
+        }
+      }
       throw new SolrException(SolrException.ErrorCode.BAD_REQUEST,
                               "Error CREATEing SolrCore '" + name + "': " +
                               ex.getMessage(), ex);

solr/core/src/test/org/apache/solr/cloud/BasicDistributedZkTest.java

@@ -333,12 +333,44 @@ public class BasicDistributedZkTest extends AbstractFullDistribZkTestBase {
     testUpdateProcessorsRunOnlyOnce("distrib-dup-test-chain-implicit");
 
     testStopAndStartCoresInOneInstance();
+    testFailedCoreCreateCleansUp();
     // Thread.sleep(10000000000L);
     if (DEBUG) {
       super.printLayout();
     }
   }
   
+  private void testFailedCoreCreateCleansUp() throws Exception {
+    Create createCmd = new Create();
+    createCmd.setCoreName("core1");
+    createCmd.setCollection("the_core_collection");
+    String coredataDir = dataDir.getAbsolutePath() + File.separator
+        + System.currentTimeMillis() + "the_core_collection";
+    createCmd.setDataDir(coredataDir);
+    createCmd.setNumShards(1);
+    createCmd.setSchemaName("nonexistent_schema.xml");
+    
+    String url = getBaseUrl(clients.get(0));
+    final HttpSolrServer server = new HttpSolrServer(url);
+    try {
+      server.request(createCmd);
+      fail("Expected SolrCore create to fail");
+    } catch (Exception e) {
+      
+    }
+    
+    long timeout = System.currentTimeMillis() + 15000;
+    while (cloudClient.getZkStateReader().getZkClient().exists("/collections/the_core_collection", true)) {
+      if (timeout <= System.currentTimeMillis()) {
+        fail(cloudClient.getZkStateReader().getZkClient().getChildren("/collections", null, true).toString() + " Collection zk node still exists");
+      }
+      Thread.sleep(100);
+    }
+    
+    
+    assertFalse("Collection zk node still exists", cloudClient.getZkStateReader().getZkClient().exists("/collections/the_core_collection", true));
+  }
+  
   private void testShardParamVariations() throws Exception {
     SolrQuery query = new SolrQuery("*:*");
     Map<String,Long> shardCounts = new HashMap<String,Long>();