HBASE-16399 Provide an API to get list of failed regions and servername in Canary (Vishal Khandelwal)

hbase-server/src/main/java/org/apache/hadoop/hbase/tool/Canary.java

@@ -39,6 +39,7 @@ import java.util.Random;
 import java.util.Set;
 import java.util.TreeSet;
 import java.util.concurrent.Callable;
+import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.ExecutionException;
 import java.util.concurrent.ExecutorService;
 import java.util.concurrent.Future;
@@ -115,11 +116,15 @@ public final class Canary implements Tool {
     public long incReadFailureCount();
     public void publishReadFailure(HRegionInfo region, Exception e);
     public void publishReadFailure(HRegionInfo region, HColumnDescriptor column, Exception e);
+    public void updateReadFailedHostList(HRegionInfo region, String serverName);
+    public Map<String,String> getReadFailures();
     public void publishReadTiming(HRegionInfo region, HColumnDescriptor column, long msTime);
     public long getWriteFailureCount();
     public void publishWriteFailure(HRegionInfo region, Exception e);
     public void publishWriteFailure(HRegionInfo region, HColumnDescriptor column, Exception e);
     public void publishWriteTiming(HRegionInfo region, HColumnDescriptor column, long msTime);
+    public void updateWriteFailedHostList(HRegionInfo region, String serverName);
+    public Map<String,String> getWriteFailures();
   }
   // new extended sink for output regionserver mode info
   // do not change the Sink interface directly due to maintaining the API
@@ -134,6 +139,9 @@ public final class Canary implements Tool {
     private AtomicLong readFailureCount = new AtomicLong(0),
         writeFailureCount = new AtomicLong(0);
 
+    private Map<String, String> readFailures = new ConcurrentHashMap<String, String>();
+    private Map<String, String> writeFailures = new ConcurrentHashMap<String, String>();
+
     @Override
     public long getReadFailureCount() {
       return readFailureCount.get();
@@ -157,12 +165,27 @@ public final class Canary implements Tool {
                 region.getRegionNameAsString(), column.getNameAsString()), e);
     }
 
+    @Override
+    public void updateReadFailedHostList(HRegionInfo region, String serverName) {
+      readFailures.put(region.getRegionNameAsString(), serverName);
+    }
+
     @Override
     public void publishReadTiming(HRegionInfo region, HColumnDescriptor column, long msTime) {
       LOG.info(String.format("read from region %s column family %s in %dms",
         region.getRegionNameAsString(), column.getNameAsString(), msTime));
     }
 
+    @Override
+    public Map<String, String> getReadFailures() {
+      return readFailures;
+    }
+
+    @Override
+    public Map<String, String> getWriteFailures() {
+      return writeFailures;
+    }
+
     @Override
     public long getWriteFailureCount() {
       return writeFailureCount.get();
@@ -186,6 +209,12 @@ public final class Canary implements Tool {
       LOG.info(String.format("write to region %s column family %s in %dms",
         region.getRegionNameAsString(), column.getNameAsString(), msTime));
     }
+
+    @Override
+    public void updateWriteFailedHostList(HRegionInfo region, String serverName) {
+      writeFailures.put(region.getRegionNameAsString(), serverName);
+    }
+
   }
   // a ExtendedSink implementation
   public static class RegionServerStdOutSink extends StdOutSink implements ExtendedSink {
@@ -265,11 +294,13 @@ public final class Canary implements Tool {
     private Sink sink;
     private TaskType taskType;
     private boolean rawScanEnabled;
+    private ServerName serverName;
 
-    RegionTask(Connection connection, HRegionInfo region, Sink sink, TaskType taskType,
+    RegionTask(Connection connection, HRegionInfo region, ServerName serverName, Sink sink,
-        boolean rawScanEnabled) {
+        TaskType taskType, boolean rawScanEnabled) {
       this.connection = connection;
       this.region = region;
+      this.serverName = serverName;
       this.sink = sink;
       this.taskType = taskType;
       this.rawScanEnabled = rawScanEnabled;
@@ -356,6 +387,7 @@ public final class Canary implements Tool {
           sink.publishReadTiming(region, column, stopWatch.getTime());
         } catch (Exception e) {
           sink.publishReadFailure(region, column, e);
+          sink.updateReadFailedHostList(region, serverName.getHostname());
         } finally {
           if (rs != null) {
             rs.close();
@@ -412,6 +444,7 @@ public final class Canary implements Tool {
         table.close();
       } catch (IOException e) {
         sink.publishWriteFailure(region, e);
+        sink.updateWriteFailedHostList(region, serverName.getHostname());
       }
       return null;
     }
@@ -733,6 +766,14 @@ public final class Canary implements Tool {
     return monitor.errorCode;
   }
 
+  public Map<String, String> getReadFailures()  {
+    return sink.getReadFailures();
+  }
+
+  public Map<String, String> getWriteFailures()  {
+    return sink.getWriteFailures();
+  }
+
   private void printUsageAndExit() {
     System.err.printf(
       "Usage: bin/hbase %s [opts] [table1 [table2]...] | [regionserver1 [regionserver2]..]%n",
@@ -1143,16 +1184,25 @@ public final class Canary implements Tool {
     } catch (TableNotFoundException e) {
       return new ArrayList<Future<Void>>();
     }
-    List<RegionTask> tasks = new ArrayList<RegionTask>();
+    finally {
-    try {
+      if (table !=null) {
-      List<HRegionInfo> hris = admin.getTableRegions(tableDesc.getTableName());
+        table.close();
-      if (hris != null) {
-        for (HRegionInfo region : hris) {
-          tasks.add(new RegionTask(admin.getConnection(), region, sink, taskType, rawScanEnabled));
       }
     }
+
+    List<RegionTask> tasks = new ArrayList<RegionTask>();
+    RegionLocator regionLocator = null;
+    try {
+      regionLocator = admin.getConnection().getRegionLocator(tableDesc.getTableName());
+      for (HRegionLocation location : regionLocator.getAllRegionLocations()) {
+        ServerName rs = location.getServerName();
+        HRegionInfo region = location.getRegionInfo();
+        tasks.add(new RegionTask(admin.getConnection(), region, rs, sink, taskType, rawScanEnabled));
+      }
     } finally {
-      table.close();
+      if (regionLocator != null) {
+        regionLocator.close();
+      }
     }
     return executor.invokeAll(tasks);
   }

hbase-server/src/test/java/org/apache/hadoop/hbase/tool/TestCanaryTool.java

@@ -53,6 +53,7 @@ import static org.mockito.Mockito.spy;
 import static org.mockito.Mockito.verify;
 import static org.mockito.Matchers.argThat;
 import static org.mockito.Mockito.never;
+import static org.junit.Assert.assertEquals;
 
 @RunWith(MockitoJUnitRunner.class)
 @Category({MediumTests.class})
@@ -110,10 +111,11 @@ public class TestCanaryTool {
     ExecutorService executor = new ScheduledThreadPoolExecutor(1);
     Canary.RegionServerStdOutSink sink = spy(new Canary.RegionServerStdOutSink());
     Canary canary = new Canary(executor, sink);
-    String[] args = { "-t", "10000", "testTable" };
+    String[] args = { "-writeSniffing", "-t", "10000", "testTable" };
     ToolRunner.run(testingUtility.getConfiguration(), canary, args);
-    verify(sink, atLeastOnce())
+    assertEquals("verify no read error count", 0, canary.getReadFailures().size());
-        .publishReadTiming(isA(HRegionInfo.class), isA(HColumnDescriptor.class), anyLong());
+    assertEquals("verify no write error count", 0, canary.getWriteFailures().size());
+    verify(sink, atLeastOnce()).publishReadTiming(isA(HRegionInfo.class), isA(HColumnDescriptor.class), anyLong());
   }
 
   //no table created, so there should be no regions
@@ -162,6 +164,7 @@ public class TestCanaryTool {
     ToolRunner.run(conf, canary, args);
     verify(sink, atLeastOnce())
         .publishReadTiming(isA(HRegionInfo.class), isA(HColumnDescriptor.class), anyLong());
+    assertEquals("verify no read error count", 0, canary.getReadFailures().size());
   }
 
   private void runRegionserverCanary() throws Exception {
@@ -169,6 +172,7 @@ public class TestCanaryTool {
     Canary canary = new Canary(executor, new Canary.RegionServerStdOutSink());
     String[] args = { "-t", "10000", "-regionserver"};
     ToolRunner.run(testingUtility.getConfiguration(), canary, args);
+    assertEquals("verify no read error count", 0, canary.getReadFailures().size());
   }
 
 }