HBASE-1928 ROOT and META tables stay in transition state (making the system not usable) if the designated regionServer dies before the assignment is complete

git-svn-id: https://svn.apache.org/repos/asf/hadoop/hbase/trunk@833779 13f79535-47bb-0310-9956-ffa450edef68

CHANGES.txt

@@ -97,6 +97,9 @@ Release 0.21.0 - Unreleased
    HBASE-1949  KeyValue expiration by Time-to-Live during major compaction is
                broken (Gary Helmling via Stack)
    HBASE-1957  Get-s can't set a Filter
+   HBASE-1928  ROOT and META tables stay in transition state (making the system
+               not usable) if the designated regionServer dies before the
+               assignment is complete (Yannis Pavlidis via Stack)
 
   IMPROVEMENTS
    HBASE-1760  Cleanup TODOs in HTable

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

@@ -86,6 +86,14 @@ class ProcessServerShutdown extends RegionServerOperation {
     if (isRootServer) {
       master.getRegionManager().unsetRootRegion();
     }
+    else {
+        //HBASE-1928: Check whether this server has been transitioning the ROOT table
+        isRootServer = master.getRegionManager().isRootServerCandidate (deadServer);
+        if (isRootServer) {
+          master.getRegionManager().unsetRootRegion();
+        }
+    }
+
     List<byte[]> metaStarts = master.getRegionManager().listMetaRegionsForServer(deadServerAddress);
 
     metaRegions = new ArrayList<MetaRegion>();
@@ -93,6 +101,12 @@ class ProcessServerShutdown extends RegionServerOperation {
       MetaRegion r = master.getRegionManager().offlineMetaRegion(region);
       metaRegions.add(r);
     }
+
+    //HBASE-1928: Check whether this server has been transitioning the META table
+    HRegionInfo metaServerRegionInfo = master.getRegionManager().getMetaServerRegionInfo (deadServer);
+    if (metaServerRegionInfo != null) {
+      metaRegions.add (new MetaRegion (deadServerAddress, metaServerRegionInfo));
+    }
   }
 
   @Override
@@ -347,4 +361,4 @@ class ProcessServerShutdown extends RegionServerOperation {
   protected int getPriority() {
     return 2; // high but not highest priority
   }
-}
+}

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

@@ -824,6 +824,44 @@ public class RegionManager implements HConstants {
     return false;
   }
 
+  /**
+   * Is this server assigned to transition the ROOT table. HBASE-1928
+   *
+   * @param server Server
+   * @return true if server is transitioning the ROOT table
+   */
+  public boolean isRootServerCandidate(final String server) {
+    for (RegionState s : regionsInTransition.values()) {
+      if (s.getRegionInfo().isRootRegion()
+          && !s.isUnassigned()
+          && s.getServerName() != null
+          && s.getServerName().equals(server)) {
+        // Has an outstanding root region to be assigned.
+        return true;
+      }
+    }
+    return false;
+  }
+
+  /**
+   * Is this server assigned to transition a META table. HBASE-1928
+   *
+   * @param server Server
+   * @return if this server was transitioning a META table then a not null HRegionInfo pointing to it
+   */
+  public HRegionInfo getMetaServerRegionInfo(final String server) {
+    for (RegionState s : regionsInTransition.values()) {
+      if (s.getRegionInfo().isMetaRegion()
+          && !s.isUnassigned()
+          && s.getServerName() != null
+          && s.getServerName().equals(server)) {
+        // Has an outstanding meta region to be assigned.
+        return s.getRegionInfo();
+      }
+    }
+    return null;
+  }
+
   /**
    * Call to take this metaserver offline for immediate reassignment.  Used only
    * when we know a region has shut down cleanly.
@@ -1604,4 +1642,4 @@ public class RegionManager implements HConstants {
       return Bytes.compareTo(getRegionName(), o.getRegionName());
     }
   }
-}
+}

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

@@ -636,6 +636,19 @@ public class ServerManager implements HConstants {
     if (info != null) {
       LOG.info("Removing server's info " + serverName);
       this.master.getRegionManager().offlineMetaServer(info.getServerAddress());
+
+      //HBASE-1928: Check whether this server has been transitioning the ROOT table
+      if (this.master.getRegionManager().isRootServerCandidate (serverName)) {
+         this.master.getRegionManager().unsetRootRegion();
+         this.master.getRegionManager().reassignRootRegion();
+      }
+
+      //HBASE-1928: Check whether this server has been transitioning the META table
+      HRegionInfo metaServerRegionInfo = this.master.getRegionManager().getMetaServerRegionInfo (serverName);
+      if (metaServerRegionInfo != null) {
+         this.master.getRegionManager().setUnassigned(metaServerRegionInfo, true);
+      }
+
       infoUpdated = true;
       // update load information
       updateLoadToServers(serverName, this.serversToLoad.remove(serverName));
@@ -831,4 +844,4 @@ public class ServerManager implements HConstants {
       m.putAll(this.loadToServers.headMap(l));
     }
   }
-}
+}