Examples with ReplicationLogCleaner org.apache.hadoop.hbase.replication.master.ReplicationLogCleaner used on opensource projects

Example 1 with ReplicationLogCleaner

use of org.apache.hadoop.hbase.replication.master.ReplicationLogCleaner in project hbase by apache.

the class TestLogsCleaner method testZnodeCversionChange.

@Test(timeout = 5000)
public void testZnodeCversionChange() throws Exception {
    Configuration conf = TEST_UTIL.getConfiguration();
    ReplicationLogCleaner cleaner = new ReplicationLogCleaner();
    cleaner.setConf(conf);
    ReplicationQueuesClientZKImpl rqcMock = Mockito.mock(ReplicationQueuesClientZKImpl.class);
    Mockito.when(rqcMock.getQueuesZNodeCversion()).thenReturn(1, 2, 3, 4);
    Field rqc = ReplicationLogCleaner.class.getDeclaredField("replicationQueues");
    rqc.setAccessible(true);
    rqc.set(cleaner, rqcMock);
    // This should return eventually when cversion stabilizes
    cleaner.getDeletableFiles(new LinkedList<>());
}

Example 2 with ReplicationLogCleaner

use of org.apache.hadoop.hbase.replication.master.ReplicationLogCleaner in project hbase by apache.

the class TestLogsCleaner method testLogCleaning.

/**
 * This tests verifies LogCleaner works correctly with WALs and Procedure WALs located
 * in the same oldWALs directory.
 * Created files:
 * - 2 invalid files
 * - 5 old Procedure WALs
 * - 30 old WALs from which 3 are in replication
 * - 5 recent Procedure WALs
 * - 1 recent WAL
 * - 1 very new WAL (timestamp in future)
 * - masterProcedureWALs subdirectory
 * Files which should stay:
 * - 3 replication WALs
 * - 2 new WALs
 * - 5 latest Procedure WALs
 * - masterProcedureWALs subdirectory
 */
@Test
public void testLogCleaning() throws Exception {
    // set TTLs
    long ttlWAL = 2000;
    long ttlProcedureWAL = 4000;
    conf.setLong("hbase.master.logcleaner.ttl", ttlWAL);
    conf.setLong("hbase.master.procedurewalcleaner.ttl", ttlProcedureWAL);
    HMaster.decorateMasterConfiguration(conf);
    Server server = new DummyServer();
    ReplicationQueueStorage queueStorage = ReplicationStorageFactory.getReplicationQueueStorage(server.getZooKeeper(), conf);
    String fakeMachineName = URLEncoder.encode(server.getServerName().toString(), StandardCharsets.UTF_8.name());
    final FileSystem fs = FileSystem.get(conf);
    fs.mkdirs(OLD_PROCEDURE_WALS_DIR);
    final long now = EnvironmentEdgeManager.currentTime();
    // Case 1: 2 invalid files, which would be deleted directly
    fs.createNewFile(new Path(OLD_WALS_DIR, "a"));
    fs.createNewFile(new Path(OLD_WALS_DIR, fakeMachineName + "." + "a"));
    // Case 2: 5 Procedure WALs that are old which would be deleted
    for (int i = 1; i <= 5; i++) {
        final Path fileName = new Path(OLD_PROCEDURE_WALS_DIR, String.format("pv2-%020d.log", i));
        fs.createNewFile(fileName);
    }
    // Sleep for sometime to get old procedure WALs
    Thread.sleep(ttlProcedureWAL - ttlWAL);
    // Case 3: old WALs which would be deletable
    for (int i = 1; i <= 30; i++) {
        Path fileName = new Path(OLD_WALS_DIR, fakeMachineName + "." + (now - i));
        fs.createNewFile(fileName);
        // files would pass TimeToLiveLogCleaner but would be rejected by ReplicationLogCleaner
        if (i % (30 / 3) == 0) {
            queueStorage.addWAL(server.getServerName(), fakeMachineName, fileName.getName());
            LOG.info("Replication log file: " + fileName);
        }
    }
    // Case 5: 5 Procedure WALs that are new, will stay
    for (int i = 6; i <= 10; i++) {
        Path fileName = new Path(OLD_PROCEDURE_WALS_DIR, String.format("pv2-%020d.log", i));
        fs.createNewFile(fileName);
    }
    // Sleep for sometime to get newer modification time
    Thread.sleep(ttlWAL);
    fs.createNewFile(new Path(OLD_WALS_DIR, fakeMachineName + "." + now));
    // Case 6: 1 newer WAL, not even deletable for TimeToLiveLogCleaner,
    // so we are not going down the chain
    fs.createNewFile(new Path(OLD_WALS_DIR, fakeMachineName + "." + (now + ttlWAL)));
    FileStatus[] status = fs.listStatus(OLD_WALS_DIR);
    LOG.info("File status: {}", Arrays.toString(status));
    // There should be 34 files and 1 masterProcedureWALs directory
    assertEquals(35, fs.listStatus(OLD_WALS_DIR).length);
    // 10 procedure WALs
    assertEquals(10, fs.listStatus(OLD_PROCEDURE_WALS_DIR).length);
    LogCleaner cleaner = new LogCleaner(1000, server, conf, fs, OLD_WALS_DIR, POOL, null);
    cleaner.chore();
    // In oldWALs we end up with the current WAL, a newer WAL, the 3 old WALs which
    // are scheduled for replication and masterProcedureWALs directory
    TEST_UTIL.waitFor(1000, (Waiter.Predicate<Exception>) () -> 6 == fs.listStatus(OLD_WALS_DIR).length);
    // In masterProcedureWALs we end up with 5 newer Procedure WALs
    TEST_UTIL.waitFor(1000, (Waiter.Predicate<Exception>) () -> 5 == fs.listStatus(OLD_PROCEDURE_WALS_DIR).length);
    if (LOG.isDebugEnabled()) {
        FileStatus[] statusOldWALs = fs.listStatus(OLD_WALS_DIR);
        FileStatus[] statusProcedureWALs = fs.listStatus(OLD_PROCEDURE_WALS_DIR);
        LOG.debug("Kept log file for oldWALs: {}", Arrays.toString(statusOldWALs));
        LOG.debug("Kept log file for masterProcedureWALs: {}", Arrays.toString(statusProcedureWALs));
    }
}

Example 3 with ReplicationLogCleaner

use of org.apache.hadoop.hbase.replication.master.ReplicationLogCleaner in project hbase by apache.

the class TestLogsCleaner method testZooKeeperNormal.

/**
 * When zk is working both files should be returned
 * @throws Exception from ZK watcher
 */
@Test
public void testZooKeeperNormal() throws Exception {
    ReplicationLogCleaner cleaner = new ReplicationLogCleaner();
    // Subtract 1000 from current time so modtime is for sure older
    // than 'now'.
    long modTime = EnvironmentEdgeManager.currentTime() - 1000;
    List<FileStatus> dummyFiles = Arrays.asList(new FileStatus(100, false, 3, 100, modTime, new Path("log1")), new FileStatus(100, false, 3, 100, modTime, new Path("log2")));
    ZKWatcher zkw = new ZKWatcher(conf, "testZooKeeperAbort-normal", null);
    try {
        cleaner.setConf(conf, zkw);
        cleaner.preClean();
        Iterable<FileStatus> filesToDelete = cleaner.getDeletableFiles(dummyFiles);
        Iterator<FileStatus> iter = filesToDelete.iterator();
        assertTrue(iter.hasNext());
        assertEquals(new Path("log1"), iter.next().getPath());
        assertTrue(iter.hasNext());
        assertEquals(new Path("log2"), iter.next().getPath());
        assertFalse(iter.hasNext());
    } finally {
        zkw.close();
    }
}

Example 4 with ReplicationLogCleaner

use of org.apache.hadoop.hbase.replication.master.ReplicationLogCleaner in project hbase by apache.

the class TestLogsCleaner method testZooKeeperAbort.

/**
   * ReplicationLogCleaner should be able to ride over ZooKeeper errors without
   * aborting.
   */
@Test
public void testZooKeeperAbort() throws Exception {
    Configuration conf = TEST_UTIL.getConfiguration();
    ReplicationLogCleaner cleaner = new ReplicationLogCleaner();
    List<FileStatus> dummyFiles = Lists.newArrayList(new FileStatus(100, false, 3, 100, System.currentTimeMillis(), new Path("log1")), new FileStatus(100, false, 3, 100, System.currentTimeMillis(), new Path("log2")));
    FaultyZooKeeperWatcher faultyZK = new FaultyZooKeeperWatcher(conf, "testZooKeeperAbort-faulty", null);
    try {
        faultyZK.init();
        cleaner.setConf(conf, faultyZK);
        // should keep all files due to a ConnectionLossException getting the queues znodes
        Iterable<FileStatus> toDelete = cleaner.getDeletableFiles(dummyFiles);
        assertFalse(toDelete.iterator().hasNext());
        assertFalse(cleaner.isStopped());
    } finally {
        faultyZK.close();
    }
    // when zk is working both files should be returned
    cleaner = new ReplicationLogCleaner();
    ZooKeeperWatcher zkw = new ZooKeeperWatcher(conf, "testZooKeeperAbort-normal", null);
    try {
        cleaner.setConf(conf, zkw);
        Iterable<FileStatus> filesToDelete = cleaner.getDeletableFiles(dummyFiles);
        Iterator<FileStatus> iter = filesToDelete.iterator();
        assertTrue(iter.hasNext());
        assertEquals(new Path("log1"), iter.next().getPath());
        assertTrue(iter.hasNext());
        assertEquals(new Path("log2"), iter.next().getPath());
        assertFalse(iter.hasNext());
    } finally {
        zkw.close();
    }
}

Example 5 with ReplicationLogCleaner

use of org.apache.hadoop.hbase.replication.master.ReplicationLogCleaner in project hbase by apache.

the class TestLogsCleaner method testZooKeeperRecoveryDuringGetListOfReplicators.

@Test
public void testZooKeeperRecoveryDuringGetListOfReplicators() throws Exception {
    ReplicationLogCleaner cleaner = new ReplicationLogCleaner();
    List<FileStatus> dummyFiles = Arrays.asList(new FileStatus(100, false, 3, 100, EnvironmentEdgeManager.currentTime(), new Path("log1")), new FileStatus(100, false, 3, 100, EnvironmentEdgeManager.currentTime(), new Path("log2")));
    FaultyZooKeeperWatcher faultyZK = new FaultyZooKeeperWatcher(conf, "testZooKeeperAbort-faulty", null);
    final AtomicBoolean getListOfReplicatorsFailed = new AtomicBoolean(false);
    try {
        faultyZK.init(false);
        ReplicationQueueStorage queueStorage = spy(ReplicationStorageFactory.getReplicationQueueStorage(faultyZK, conf));
        doAnswer(new Answer<Object>() {

            @Override
            public Object answer(InvocationOnMock invocation) throws Throwable {
                try {
                    return invocation.callRealMethod();
                } catch (ReplicationException e) {
                    LOG.debug("Caught Exception", e);
                    getListOfReplicatorsFailed.set(true);
                    throw e;
                }
            }
        }).when(queueStorage).getAllWALs();
        cleaner.setConf(conf, faultyZK, queueStorage);
        // should keep all files due to a ConnectionLossException getting the queues znodes
        cleaner.preClean();
        Iterable<FileStatus> toDelete = cleaner.getDeletableFiles(dummyFiles);
        assertTrue(getListOfReplicatorsFailed.get());
        assertFalse(toDelete.iterator().hasNext());
        assertFalse(cleaner.isStopped());
        // zk recovery.
        faultyZK.init(true);
        cleaner.preClean();
        Iterable<FileStatus> filesToDelete = cleaner.getDeletableFiles(dummyFiles);
        Iterator<FileStatus> iter = filesToDelete.iterator();
        assertTrue(iter.hasNext());
        assertEquals(new Path("log1"), iter.next().getPath());
        assertTrue(iter.hasNext());
        assertEquals(new Path("log2"), iter.next().getPath());
        assertFalse(iter.hasNext());
    } finally {
        faultyZK.close();
    }
}