HBASE-490 Doubly-assigned .META.; master uses one and clients another

git-svn-id: https://svn.apache.org/repos/asf/hadoop/hbase/trunk@633597 13f79535-47bb-0310-9956-ffa450edef68

CHANGES.txt

@@ -28,6 +28,7 @@ Hbase Change Log
    HBASE-462   Update migration tool
    HBASE-473   When a table is deleted, master sends multiple close messages to
                the region server
+   HBASE-490   Doubly-assigned .META.; master uses one and clients another
    
   IMPROVEMENTS
    HBASE-415   Rewrite leases to use DelayedBlockingQueue instead of polling

src/java/org/apache/hadoop/hbase/master/BaseScanner.java

@@ -372,21 +372,13 @@ abstract class BaseScanner extends Chore implements HConstants {
     }
 
     /*
-     * If the server is not dead and either:
-     *   the stored info is not null and the start code does not match
-     * or:
-     *   the stored info is null and the region is neither unassigned nor pending
-     * then:
+     * If the server is a dead server or its startcode is off -- either null
+     * or doesn't match the start code for the address -- then add it to the
+     * list of unassigned regions IF not already there (or pending open).
      */ 
-    if (!deadServer &&
-      ((storedInfo != null && storedInfo.getStartCode() != startCode) ||
-        (storedInfo == null &&
-          !regionManager.isUnassigned(info) &&
+    if (!deadServer && !regionManager.isUnassigned(info) &&
           !regionManager.isPending(info.getRegionName())
-        )
-      )
-    ) {
-
+        && (storedInfo == null || storedInfo.getStartCode() != startCode)) {
       // The current assignment is invalid
       if (LOG.isDebugEnabled()) {
         LOG.debug("Current assignment of " + info.getRegionName() +

src/java/org/apache/hadoop/hbase/master/MetaScanner.java

@@ -38,7 +38,7 @@ import org.apache.hadoop.hbase.RemoteExceptionHandler;
  * action would prevent other work from getting done.
  */
 class MetaScanner extends BaseScanner {
-  /** Work for the meta scanner is queued up here */
+  /** Initial work for the meta scanner is queued up here */
   private volatile BlockingQueue<MetaRegion> metaRegionsToScan =
     new LinkedBlockingQueue<MetaRegion>();
     
@@ -50,10 +50,10 @@ class MetaScanner extends BaseScanner {
     super(master, regionManager, false, master.metaRescanInterval, master.closed);
   }
 
+  // Don't retry if we get an error while scanning. Errors are most often
+  // caused by the server going away. Wait until next rescan interval when
+  // things should be back to normal
   private boolean scanOneMetaRegion(MetaRegion region) {
-    // Don't retry if we get an error while scanning. Errors are most often
-    // caused by the server going away. Wait until next rescan interval when
-    // things should be back to normal
     boolean scanSuccessful = false;
     while (!master.closed.get() && !regionManager.isInitialRootScanComplete() &&
       regionManager.getRootRegionLocation() == null) {
@@ -95,7 +95,9 @@ class MetaScanner extends BaseScanner {
   @Override
   protected boolean initialScan() {
     MetaRegion region = null;
-    while (!master.closed.get() && region == null && !metaRegionsScanned()) {
+    while (!master.closed.get() &&
+        (region == null && metaRegionsToScan.size() > 0) &&
+        !metaRegionsScanned()) {
       try {
         region = metaRegionsToScan.poll(master.threadWakeFrequency, 
           TimeUnit.MILLISECONDS);
@@ -164,4 +166,4 @@ class MetaScanner extends BaseScanner {
   void addMetaRegionToScan(MetaRegion m) throws InterruptedException {
     metaRegionsToScan.add(m);
   }
-}
+}

src/java/org/apache/hadoop/hbase/master/RegionManager.java

@@ -598,7 +598,7 @@ class RegionManager implements HConstants {
   
   public void waitForRootRegionLocation() {
     synchronized (rootRegionLocation) {
-      while(!master.closed.get() && rootRegionLocation.get() == null) {
+      while (!master.closed.get() && rootRegionLocation.get() == null) {
         // rootRegionLocation will be filled in when we get an 'open region'
         // regionServerReport message from the HRegionServer that has been
         // allocated the ROOT region below.

src/java/org/apache/hadoop/hbase/master/RootScanner.java

@@ -31,10 +31,10 @@ class RootScanner extends BaseScanner {
     super(master, regionManager, true, master.metaRescanInterval, master.closed);
   }
 
+  // Don't retry if we get an error while scanning. Errors are most often
+  // caused by the server going away. Wait until next rescan interval when
+  // things should be back to normal
   private boolean scanRoot() {
-    // Don't retry if we get an error while scanning. Errors are most often
-    // caused by the server going away. Wait until next rescan interval when
-    // things should be back to normal
     boolean scanSuccessful = false;
     master.waitForRootRegionLocation();
     if (master.closed.get()) {
@@ -71,4 +71,4 @@ class RootScanner extends BaseScanner {
   protected void maintenanceScan() {
     scanRoot();
   }
-}
+}

src/java/org/apache/hadoop/hbase/util/Sleeper.java

@@ -44,7 +44,7 @@ public class Sleeper {
    * Sleep for period.
    */
   public void sleep() {
-    sleep(period);
+    sleep(System.currentTimeMillis());
   }
   
   /**
@@ -69,4 +69,4 @@ public class Sleeper {
       }
     }
   }
-}
+}