Examples with ZooKeeperWatcher org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher used on opensource projects

Example 41 with ZooKeeperWatcher

use of org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher in project hbase by apache.

the class TestDistributedLogSplitting method testThreeRSAbort.

@Test(timeout = 300000)
public void testThreeRSAbort() throws Exception {
    LOG.info("testThreeRSAbort");
    final int NUM_REGIONS_TO_CREATE = 40;
    final int NUM_ROWS_PER_REGION = 100;
    // NUM_RS=6.
    startCluster(NUM_RS);
    final ZooKeeperWatcher zkw = new ZooKeeperWatcher(conf, "distributed log splitting test", null);
    Table ht = installTable(zkw, "table", "family", NUM_REGIONS_TO_CREATE);
    try {
        populateDataInTable(NUM_ROWS_PER_REGION, "family");
        List<RegionServerThread> rsts = cluster.getLiveRegionServerThreads();
        assertEquals(NUM_RS, rsts.size());
        rsts.get(0).getRegionServer().abort("testing");
        rsts.get(1).getRegionServer().abort("testing");
        rsts.get(2).getRegionServer().abort("testing");
        long start = EnvironmentEdgeManager.currentTime();
        while (cluster.getLiveRegionServerThreads().size() > (NUM_RS - 3)) {
            if (EnvironmentEdgeManager.currentTime() - start > 60000) {
                assertTrue(false);
            }
            Thread.sleep(200);
        }
        start = EnvironmentEdgeManager.currentTime();
        while (HBaseTestingUtility.getAllOnlineRegions(cluster).size() < (NUM_REGIONS_TO_CREATE + 1)) {
            if (EnvironmentEdgeManager.currentTime() - start > 60000) {
                assertTrue("Timedout", false);
            }
            Thread.sleep(200);
        }
        // wait for all regions are fully recovered
        TEST_UTIL.waitFor(180000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                List<String> recoveringRegions = zkw.getRecoverableZooKeeper().getChildren(zkw.znodePaths.recoveringRegionsZNode, false);
                return (recoveringRegions != null && recoveringRegions.isEmpty());
            }
        });
        assertEquals(NUM_REGIONS_TO_CREATE * NUM_ROWS_PER_REGION, TEST_UTIL.countRows(ht));
    } finally {
        if (ht != null)
            ht.close();
        if (zkw != null)
            zkw.close();
    }
}

Example 42 with ZooKeeperWatcher

use of org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher in project hbase by apache.

the class TestDistributedLogSplitting method testLogReplayTwoSequentialRSDown.

@Ignore("DLR is broken by HBASE-12751")
@Test(timeout = 300000)
public void testLogReplayTwoSequentialRSDown() throws Exception {
    LOG.info("testRecoveredEditsReplayTwoSequentialRSDown");
    conf.setBoolean(HConstants.DISTRIBUTED_LOG_REPLAY_KEY, true);
    startCluster(NUM_RS);
    final int NUM_REGIONS_TO_CREATE = 40;
    final int NUM_LOG_LINES = 1000;
    // turn off load balancing to prevent regions from moving around otherwise
    // they will consume recovered.edits
    master.balanceSwitch(false);
    List<RegionServerThread> rsts = cluster.getLiveRegionServerThreads();
    final ZooKeeperWatcher zkw = new ZooKeeperWatcher(conf, "table-creation", null);
    Table ht = installTable(zkw, "table", "family", NUM_REGIONS_TO_CREATE);
    try {
        List<HRegionInfo> regions = null;
        HRegionServer hrs1 = findRSToKill(false, "table");
        regions = ProtobufUtil.getOnlineRegions(hrs1.getRSRpcServices());
        makeWAL(hrs1, regions, "table", "family", NUM_LOG_LINES, 100);
        // abort RS1
        LOG.info("Aborting region server: " + hrs1.getServerName());
        hrs1.abort("testing");
        // wait for abort completes
        TEST_UTIL.waitFor(120000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return (cluster.getLiveRegionServerThreads().size() <= (NUM_RS - 1));
            }
        });
        // wait for regions come online
        TEST_UTIL.waitFor(180000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return (HBaseTestingUtility.getAllOnlineRegions(cluster).size() >= (NUM_REGIONS_TO_CREATE + 1));
            }
        });
        // sleep a little bit in order to interrupt recovering in the middle
        Thread.sleep(300);
        // abort second region server
        rsts = cluster.getLiveRegionServerThreads();
        HRegionServer hrs2 = rsts.get(0).getRegionServer();
        LOG.info("Aborting one more region server: " + hrs2.getServerName());
        hrs2.abort("testing");
        // wait for abort completes
        TEST_UTIL.waitFor(120000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return (cluster.getLiveRegionServerThreads().size() <= (NUM_RS - 2));
            }
        });
        // wait for regions come online
        TEST_UTIL.waitFor(180000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return (HBaseTestingUtility.getAllOnlineRegions(cluster).size() >= (NUM_REGIONS_TO_CREATE + 1));
            }
        });
        // wait for all regions are fully recovered
        TEST_UTIL.waitFor(180000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                List<String> recoveringRegions = zkw.getRecoverableZooKeeper().getChildren(zkw.znodePaths.recoveringRegionsZNode, false);
                return (recoveringRegions != null && recoveringRegions.isEmpty());
            }
        });
        assertEquals(NUM_LOG_LINES, TEST_UTIL.countRows(ht));
    } finally {
        if (ht != null)
            ht.close();
        if (zkw != null)
            zkw.close();
    }
}

Example 43 with ZooKeeperWatcher

use of org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher in project hbase by apache.

the class TestDistributedLogSplitting method testMasterStartsUpWithLogReplayWork.

@Ignore("DLR is broken by HBASE-12751")
@Test(timeout = 300000)
public void testMasterStartsUpWithLogReplayWork() throws Exception {
    LOG.info("testMasterStartsUpWithLogReplayWork");
    conf.setBoolean(HConstants.DISTRIBUTED_LOG_REPLAY_KEY, true);
    conf.setInt(ServerManager.WAIT_ON_REGIONSERVERS_MINTOSTART, NUM_RS - 1);
    startCluster(NUM_RS);
    final int NUM_REGIONS_TO_CREATE = 40;
    final int NUM_LOG_LINES = 1000;
    // turn off load balancing to prevent regions from moving around otherwise
    // they will consume recovered.edits
    master.balanceSwitch(false);
    final ZooKeeperWatcher zkw = new ZooKeeperWatcher(conf, "table-creation", null);
    Table ht = installTable(zkw, "table", "family", NUM_REGIONS_TO_CREATE);
    try {
        HRegionServer hrs = findRSToKill(false, "table");
        List<HRegionInfo> regions = ProtobufUtil.getOnlineRegions(hrs.getRSRpcServices());
        makeWAL(hrs, regions, "table", "family", NUM_LOG_LINES, 100);
        // abort master
        abortMaster(cluster);
        // abort RS
        LOG.info("Aborting region server: " + hrs.getServerName());
        hrs.abort("testing");
        // wait for the RS dies
        TEST_UTIL.waitFor(120000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return (cluster.getLiveRegionServerThreads().size() <= (NUM_RS - 1));
            }
        });
        Thread.sleep(2000);
        LOG.info("Current Open Regions:" + HBaseTestingUtility.getAllOnlineRegions(cluster).size());
        // wait for all regions are fully recovered
        TEST_UTIL.waitFor(180000, 200, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                List<String> recoveringRegions = zkw.getRecoverableZooKeeper().getChildren(zkw.znodePaths.recoveringRegionsZNode, false);
                boolean done = recoveringRegions != null && recoveringRegions.isEmpty();
                if (!done) {
                    LOG.info("Recovering regions: " + recoveringRegions);
                }
                return done;
            }
        });
        LOG.info("Current Open Regions After Master Node Starts Up:" + HBaseTestingUtility.getAllOnlineRegions(cluster).size());
        assertEquals(NUM_LOG_LINES, TEST_UTIL.countRows(ht));
    } finally {
        if (ht != null)
            ht.close();
        if (zkw != null)
            zkw.close();
    }
}

Example 44 with ZooKeeperWatcher

use of org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher in project hbase by apache.

the class TestDistributedLogSplitting method testLogReplayWithMetaRSDown.

@Ignore("DLR is broken by HBASE-12751")
@Test(timeout = 300000)
public void testLogReplayWithMetaRSDown() throws Exception {
    LOG.info("testRecoveredEditsReplayWithMetaRSDown");
    conf.setBoolean(HConstants.DISTRIBUTED_LOG_REPLAY_KEY, true);
    startCluster(NUM_RS);
    final int NUM_REGIONS_TO_CREATE = 40;
    final int NUM_LOG_LINES = 1000;
    // turn off load balancing to prevent regions from moving around otherwise
    // they will consume recovered.edits
    master.balanceSwitch(false);
    final ZooKeeperWatcher zkw = new ZooKeeperWatcher(conf, "table-creation", null);
    Table ht = installTable(zkw, "table", "family", NUM_REGIONS_TO_CREATE);
    try {
        HRegionServer hrs = findRSToKill(true, "table");
        List<HRegionInfo> regions = ProtobufUtil.getOnlineRegions(hrs.getRSRpcServices());
        makeWAL(hrs, regions, "table", "family", NUM_LOG_LINES, 100);
        this.abortRSAndVerifyRecovery(hrs, ht, zkw, NUM_REGIONS_TO_CREATE, NUM_LOG_LINES);
    } finally {
        if (ht != null)
            ht.close();
        if (zkw != null)
            zkw.close();
    }
}

Example 45 with ZooKeeperWatcher

use of org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher in project hbase by apache.

the class TestDistributedLogSplitting method testMarkRegionsRecoveringInZK.

@Ignore("DLR is broken by HBASE-12751")
@Test(timeout = 300000)
public void testMarkRegionsRecoveringInZK() throws Exception {
    LOG.info("testMarkRegionsRecoveringInZK");
    conf.setBoolean(HConstants.DISTRIBUTED_LOG_REPLAY_KEY, true);
    startCluster(NUM_RS);
    master.balanceSwitch(false);
    List<RegionServerThread> rsts = cluster.getLiveRegionServerThreads();
    final ZooKeeperWatcher zkw = master.getZooKeeper();
    Table ht = installTable(zkw, "table", "family", 40);
    try {
        final SplitLogManager slm = master.getMasterWalManager().getSplitLogManager();
        Set<HRegionInfo> regionSet = new HashSet<>();
        HRegionInfo region = null;
        HRegionServer hrs = null;
        ServerName firstFailedServer = null;
        ServerName secondFailedServer = null;
        for (int i = 0; i < NUM_RS; i++) {
            hrs = rsts.get(i).getRegionServer();
            List<HRegionInfo> regions = ProtobufUtil.getOnlineRegions(hrs.getRSRpcServices());
            if (regions.isEmpty())
                continue;
            region = regions.get(0);
            regionSet.add(region);
            firstFailedServer = hrs.getServerName();
            secondFailedServer = rsts.get((i + 1) % NUM_RS).getRegionServer().getServerName();
            break;
        }
        slm.markRegionsRecovering(firstFailedServer, regionSet);
        slm.markRegionsRecovering(secondFailedServer, regionSet);
        List<String> recoveringRegions = ZKUtil.listChildrenNoWatch(zkw, ZKUtil.joinZNode(zkw.znodePaths.recoveringRegionsZNode, region.getEncodedName()));
        assertEquals(recoveringRegions.size(), 2);
        // wait for splitLogWorker to mark them up because there is no WAL files recorded in ZK
        final HRegionServer tmphrs = hrs;
        TEST_UTIL.waitFor(60000, 1000, new Waiter.Predicate<Exception>() {

            @Override
            public boolean evaluate() throws Exception {
                return (tmphrs.getRecoveringRegions().isEmpty());
            }
        });
    } finally {
        if (ht != null)
            ht.close();
        if (zkw != null)
            zkw.close();
    }
}