Examples with ClusterMetrics org.apache.hadoop.hbase.ClusterMetrics used on opensource projects

Example 36 with ClusterMetrics

use of org.apache.hadoop.hbase.ClusterMetrics in project hbase by apache.

the class TestReplicationStatus method testReplicationStatus.

/**
 * Test for HBASE-9531.
 * <p/>
 * put a few rows into htable1, which should be replicated to htable2 <br/>
 * create a ClusterStatus instance 'status' from HBaseAdmin <br/>
 * test : status.getLoad(server).getReplicationLoadSourceList() <br/>
 * test : status.getLoad(server).getReplicationLoadSink()
 */
@Test
public void testReplicationStatus() throws Exception {
    // This test wants two RS's up. We only run one generally so add one.
    UTIL1.getMiniHBaseCluster().startRegionServer();
    Waiter.waitFor(UTIL1.getConfiguration(), 30000, new Waiter.Predicate<Exception>() {

        @Override
        public boolean evaluate() throws Exception {
            return UTIL1.getMiniHBaseCluster().getLiveRegionServerThreads().size() > 1;
        }
    });
    Admin hbaseAdmin = UTIL1.getAdmin();
    // disable peer <= WHY? I DON'T GET THIS DISABLE BUT TEST FAILS W/O IT.
    hbaseAdmin.disableReplicationPeer(PEER_ID2);
    insertRowsOnSource();
    LOG.info("AFTER PUTS");
    // TODO: Change this wait to a barrier. I tried waiting on replication stats to
    // change but sleeping in main thread seems to mess up background replication.
    // HACK! To address flakeyness.
    Threads.sleep(10000);
    ClusterMetrics metrics = hbaseAdmin.getClusterMetrics(EnumSet.of(Option.LIVE_SERVERS));
    for (JVMClusterUtil.RegionServerThread thread : UTIL1.getHBaseCluster().getRegionServerThreads()) {
        ServerName server = thread.getRegionServer().getServerName();
        assertTrue("" + server, metrics.getLiveServerMetrics().containsKey(server));
        ServerMetrics sm = metrics.getLiveServerMetrics().get(server);
        List<ReplicationLoadSource> rLoadSourceList = sm.getReplicationLoadSourceList();
        ReplicationLoadSink rLoadSink = sm.getReplicationLoadSink();
        // check SourceList only has one entry, because only has one peer
        assertEquals("Failed to get ReplicationLoadSourceList " + rLoadSourceList + ", " + server, 1, rLoadSourceList.size());
        assertEquals(PEER_ID2, rLoadSourceList.get(0).getPeerID());
        // check Sink exist only as it is difficult to verify the value on the fly
        assertTrue("failed to get ReplicationLoadSink.AgeOfLastShippedOp ", (rLoadSink.getAgeOfLastAppliedOp() >= 0));
        assertTrue("failed to get ReplicationLoadSink.TimeStampsOfLastAppliedOp ", (rLoadSink.getTimestampsOfLastAppliedOp() >= 0));
    }
    // Stop rs1, then the queue of rs1 will be transfered to rs0
    HRegionServer hrs = UTIL1.getHBaseCluster().getRegionServer(1);
    hrs.stop("Stop RegionServer");
    while (hrs.isAlive()) {
        Threads.sleep(100);
    }
    // To be sure it dead and references cleaned up. TODO: Change this to a barrier.
    // I tried waiting on replication stats to change but sleeping in main thread
    // seems to mess up background replication.
    Threads.sleep(10000);
    ServerName server = UTIL1.getHBaseCluster().getRegionServer(0).getServerName();
    List<ReplicationLoadSource> rLoadSourceList = waitOnMetricsReport(1, server);
    // The remaining server should now have two queues -- the original and then the one that was
    // added because of failover. The original should still be PEER_ID2 though.
    assertEquals("Failed ReplicationLoadSourceList " + rLoadSourceList, 2, rLoadSourceList.size());
    assertEquals(PEER_ID2, rLoadSourceList.get(0).getPeerID());
}

Example 37 with ClusterMetrics

use of org.apache.hadoop.hbase.ClusterMetrics in project hbase by apache.

the class TestReplicationStatusSink method getLatestSinkMetric.

private ReplicationLoadSink getLatestSinkMetric(Admin admin, ServerName server) throws IOException {
    ClusterMetrics metrics = admin.getClusterMetrics(EnumSet.of(ClusterMetrics.Option.LIVE_SERVERS));
    ServerMetrics sm = metrics.getLiveServerMetrics().get(server);
    return sm.getReplicationLoadSink();
}

Example 38 with ClusterMetrics

use of org.apache.hadoop.hbase.ClusterMetrics in project hbase by apache.

the class TestRegionsRecoveryChore method testRegionReopensWithStoreRefConfig.

@Test
public void testRegionReopensWithStoreRefConfig() throws Exception {
    regionNo = 0;
    ClusterMetrics clusterMetrics = TestRegionsRecoveryChore.getClusterMetrics(4);
    final Map<ServerName, ServerMetrics> serverMetricsMap = clusterMetrics.getLiveServerMetrics();
    LOG.debug("All Region Names with refCount....");
    for (ServerMetrics serverMetrics : serverMetricsMap.values()) {
        Map<byte[], RegionMetrics> regionMetricsMap = serverMetrics.getRegionMetrics();
        for (RegionMetrics regionMetrics : regionMetricsMap.values()) {
            LOG.debug("name: " + new String(regionMetrics.getRegionName()) + " refCount: " + regionMetrics.getStoreRefCount());
        }
    }
    Mockito.when(hMaster.getClusterMetrics()).thenReturn(clusterMetrics);
    Mockito.when(hMaster.getAssignmentManager()).thenReturn(assignmentManager);
    for (byte[] regionName : REGION_NAME_LIST) {
        Mockito.when(assignmentManager.getRegionInfo(regionName)).thenReturn(TestRegionsRecoveryChore.getRegionInfo(regionName));
    }
    Stoppable stoppable = new StoppableImplementation();
    Configuration configuration = getCustomConf();
    configuration.setInt("hbase.regions.recovery.store.file.ref.count", 300);
    regionsRecoveryChore = new RegionsRecoveryChore(stoppable, configuration, hMaster);
    regionsRecoveryChore.chore();
    // Verify that we need to reopen regions of 2 tables
    Mockito.verify(hMaster, Mockito.times(2)).reopenRegions(Mockito.any(), Mockito.anyList(), Mockito.anyLong(), Mockito.anyLong());
    Mockito.verify(hMaster, Mockito.times(1)).getClusterMetrics();
    // Verify that we need to reopen total 3 regions that have refCount > 300
    Mockito.verify(hMaster, Mockito.times(3)).getAssignmentManager();
    Mockito.verify(assignmentManager, Mockito.times(3)).getRegionInfo(Mockito.any());
}

Example 39 with ClusterMetrics

use of org.apache.hadoop.hbase.ClusterMetrics in project hbase by apache.

the class TestFavoredStochasticBalancerPickers method testPickers.

@Test
public void testPickers() throws Exception {
    TableName tableName = TableName.valueOf(name.getMethodName());
    ColumnFamilyDescriptor columnFamilyDescriptor = ColumnFamilyDescriptorBuilder.newBuilder(HConstants.CATALOG_FAMILY).build();
    TableDescriptor desc = TableDescriptorBuilder.newBuilder(tableName).setColumnFamily(columnFamilyDescriptor).build();
    admin.createTable(desc, Bytes.toBytes("aaa"), Bytes.toBytes("zzz"), REGIONS);
    TEST_UTIL.waitUntilAllRegionsAssigned(tableName);
    TEST_UTIL.loadTable(admin.getConnection().getTable(tableName), HConstants.CATALOG_FAMILY);
    admin.flush(tableName);
    HMaster master = cluster.getMaster();
    FavoredNodesManager fnm = master.getFavoredNodesManager();
    ServerName masterServerName = master.getServerName();
    List<ServerName> excludedServers = Lists.newArrayList(masterServerName);
    final ServerName mostLoadedServer = getRSWithMaxRegions(tableName, excludedServers);
    assertNotNull(mostLoadedServer);
    int numRegions = getTableRegionsFromServer(tableName, mostLoadedServer).size();
    excludedServers.add(mostLoadedServer);
    // Lets find another server with more regions to calculate number of regions to move
    ServerName source = getRSWithMaxRegions(tableName, excludedServers);
    assertNotNull(source);
    int regionsToMove = getTableRegionsFromServer(tableName, source).size() / 2;
    // Since move only works if the target is part of favored nodes of the region, lets get all
    // regions that are movable to mostLoadedServer
    List<RegionInfo> hris = getRegionsThatCanBeMoved(tableName, mostLoadedServer);
    RegionStates rst = master.getAssignmentManager().getRegionStates();
    for (int i = 0; i < regionsToMove; i++) {
        final RegionInfo regionInfo = hris.get(i);
        admin.move(regionInfo.getEncodedNameAsBytes(), mostLoadedServer);
        LOG.info("Moving region: " + hris.get(i).getRegionNameAsString() + " to " + mostLoadedServer);
        TEST_UTIL.waitFor(60000, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return ServerName.isSameAddress(rst.getRegionServerOfRegion(regionInfo), mostLoadedServer);
            }
        });
    }
    final int finalRegions = numRegions + regionsToMove;
    TEST_UTIL.waitUntilNoRegionsInTransition(60000);
    TEST_UTIL.waitFor(60000, new Waiter.Predicate<Exception>() {

        @Override
        public boolean evaluate() throws Exception {
            int numRegions = getTableRegionsFromServer(tableName, mostLoadedServer).size();
            return (numRegions == finalRegions);
        }
    });
    TEST_UTIL.getHBaseCluster().startRegionServerAndWait(60000);
    Map<ServerName, List<RegionInfo>> serverAssignments = Maps.newHashMap();
    ClusterMetrics status = admin.getClusterMetrics(EnumSet.of(Option.LIVE_SERVERS));
    for (ServerName sn : status.getLiveServerMetrics().keySet()) {
        if (!ServerName.isSameAddress(sn, masterServerName)) {
            serverAssignments.put(sn, getTableRegionsFromServer(tableName, sn));
        }
    }
    RegionHDFSBlockLocationFinder regionFinder = new RegionHDFSBlockLocationFinder();
    regionFinder.setClusterMetrics(admin.getClusterMetrics(EnumSet.of(Option.LIVE_SERVERS)));
    regionFinder.setConf(conf);
    regionFinder.setClusterInfoProvider(new MasterClusterInfoProvider(TEST_UTIL.getMiniHBaseCluster().getMaster()));
    BalancerClusterState cluster = new BalancerClusterState(serverAssignments, null, regionFinder, new RackManager(conf));
    LoadOnlyFavoredStochasticBalancer balancer = (LoadOnlyFavoredStochasticBalancer) TEST_UTIL.getMiniHBaseCluster().getMaster().getLoadBalancer().getInternalBalancer();
    cluster.sortServersByRegionCount();
    Integer[] servers = cluster.serverIndicesSortedByRegionCount;
    LOG.info("Servers sorted by region count:" + Arrays.toString(servers));
    LOG.info("Cluster dump: " + cluster);
    if (!mostLoadedServer.equals(cluster.servers[servers[servers.length - 1]])) {
        LOG.error("Most loaded server: " + mostLoadedServer + " does not match: " + cluster.servers[servers[servers.length - 1]]);
    }
    assertEquals(mostLoadedServer, cluster.servers[servers[servers.length - 1]]);
    FavoredStochasticBalancer.FavoredNodeLoadPicker loadPicker = balancer.new FavoredNodeLoadPicker();
    boolean userRegionPicked = false;
    for (int i = 0; i < 100; i++) {
        if (userRegionPicked) {
            break;
        } else {
            BalanceAction action = loadPicker.generate(cluster);
            if (action.getType() == BalanceAction.Type.MOVE_REGION) {
                MoveRegionAction moveRegionAction = (MoveRegionAction) action;
                RegionInfo region = cluster.regions[moveRegionAction.getRegion()];
                assertNotEquals(-1, moveRegionAction.getToServer());
                ServerName destinationServer = cluster.servers[moveRegionAction.getToServer()];
                assertEquals(cluster.servers[moveRegionAction.getFromServer()], mostLoadedServer);
                if (!region.getTable().isSystemTable()) {
                    List<ServerName> favNodes = fnm.getFavoredNodes(region);
                    assertTrue(favNodes.contains(ServerName.valueOf(destinationServer.getAddress(), -1)));
                    userRegionPicked = true;
                }
            }
        }
    }
    assertTrue("load picker did not pick expected regions in 100 iterations.", userRegionPicked);
}