Examples with FailedLogCloseException org.apache.hadoop.hbase.regionserver.wal.FailedLogCloseException used on opensource projects

Example 1 with FailedLogCloseException

use of org.apache.hadoop.hbase.regionserver.wal.FailedLogCloseException in project hbase by apache.

the class LogRoller method run.

@Override
public void run() {
    while (running) {
        long now = System.currentTimeMillis();
        boolean periodic = false;
        if (!rollLog.get()) {
            periodic = (now - this.lastrolltime) > this.rollperiod;
            if (!periodic) {
                synchronized (rollLog) {
                    try {
                        if (!rollLog.get()) {
                            rollLog.wait(this.threadWakeFrequency);
                        }
                    } catch (InterruptedException e) {
                    // Fall through
                    }
                }
                continue;
            }
            // Time for periodic roll
            if (LOG.isDebugEnabled()) {
                LOG.debug("Wal roll period " + this.rollperiod + "ms elapsed");
            }
        } else if (LOG.isDebugEnabled()) {
            LOG.debug("WAL roll requested");
        }
        // FindBugs UL_UNRELEASED_LOCK_EXCEPTION_PATH
        rollLock.lock();
        try {
            this.lastrolltime = now;
            for (Entry<WAL, Boolean> entry : walNeedsRoll.entrySet()) {
                final WAL wal = entry.getKey();
                // Force the roll if the logroll.period is elapsed or if a roll was requested.
                // The returned value is an array of actual region names.
                final byte[][] regionsToFlush = wal.rollWriter(periodic || entry.getValue().booleanValue());
                walNeedsRoll.put(wal, Boolean.FALSE);
                if (regionsToFlush != null) {
                    for (byte[] r : regionsToFlush) scheduleFlush(r);
                }
            }
        } catch (FailedLogCloseException e) {
            server.abort("Failed log close in log roller", e);
        } catch (java.net.ConnectException e) {
            server.abort("Failed log close in log roller", e);
        } catch (IOException ex) {
            // Abort if we get here.  We probably won't recover an IOE. HBASE-1132
            server.abort("IOE in log roller", ex instanceof RemoteException ? ((RemoteException) ex).unwrapRemoteException() : ex);
        } catch (Exception ex) {
            LOG.error("Log rolling failed", ex);
            server.abort("Log rolling failed", ex);
        } finally {
            try {
                rollLog.set(false);
            } finally {
                rollLock.unlock();
            }
        }
    }
    LOG.info("LogRoller exiting.");
}

Example 2 with FailedLogCloseException

use of org.apache.hadoop.hbase.regionserver.wal.FailedLogCloseException in project hbase by apache.

the class TestWALLockup method testLockup16960.

/**
   * Reproduce locking up that happens when there's no further syncs after
   * append fails, and causing an isolated sync then infinite wait. See
   * HBASE-16960. If below is broken, we will see this test timeout because it
   * is locked up.
   * <p/>
   * Steps for reproduce:<br/>
   * 1. Trigger server abort through dodgyWAL1<br/>
   * 2. Add a {@link DummyWALActionsListener} to dodgyWAL2 to cause ringbuffer
   * event handler thread sleep for a while thus keeping {@code endOfBatch}
   * false<br/>
   * 3. Publish a sync then an append which will throw exception, check whether
   * the sync could return
   */
@Test(timeout = 20000)
public void testLockup16960() throws IOException {
    // A WAL that we can have throw exceptions when a flag is set.
    class DodgyFSLog extends FSHLog {

        // Set this when want the WAL to start throwing exceptions.
        volatile boolean throwException = false;

        public DodgyFSLog(FileSystem fs, Path root, String logDir, Configuration conf) throws IOException {
            super(fs, root, logDir, conf);
        }

        @Override
        protected Writer createWriterInstance(Path path) throws IOException {
            final Writer w = super.createWriterInstance(path);
            return new Writer() {

                @Override
                public void close() throws IOException {
                    w.close();
                }

                @Override
                public void sync() throws IOException {
                    if (throwException) {
                        throw new IOException("FAKE! Failed to replace a bad datanode...SYNC");
                    }
                    w.sync();
                }

                @Override
                public void append(Entry entry) throws IOException {
                    if (throwException) {
                        throw new IOException("FAKE! Failed to replace a bad datanode...APPEND");
                    }
                    w.append(entry);
                }

                @Override
                public long getLength() {
                    return w.getLength();
                }
            };
        }

        @Override
        protected long doReplaceWriter(Path oldPath, Path newPath, Writer nextWriter) throws IOException {
            if (throwException) {
                throw new FailedLogCloseException("oldPath=" + oldPath + ", newPath=" + newPath);
            }
            long oldFileLen = 0L;
            oldFileLen = super.doReplaceWriter(oldPath, newPath, nextWriter);
            return oldFileLen;
        }
    }
    // Mocked up server and regionserver services. Needed below.
    Server server = new DummyServer(CONF, ServerName.valueOf("hostname1.example.org", 1234, 1L).toString());
    RegionServerServices services = Mockito.mock(RegionServerServices.class);
    CONF.setLong("hbase.regionserver.hlog.sync.timeout", 10000);
    // OK. Now I have my mocked up Server & RegionServerServices and dodgy WAL,
    // go ahead with test.
    FileSystem fs = FileSystem.get(CONF);
    Path rootDir = new Path(dir + getName());
    DodgyFSLog dodgyWAL1 = new DodgyFSLog(fs, rootDir, getName(), CONF);
    Path rootDir2 = new Path(dir + getName() + "2");
    final DodgyFSLog dodgyWAL2 = new DodgyFSLog(fs, rootDir2, getName() + "2", CONF);
    // Add a listener to force ringbuffer event handler sleep for a while
    dodgyWAL2.registerWALActionsListener(new DummyWALActionsListener());
    // I need a log roller running.
    LogRoller logRoller = new LogRoller(server, services);
    logRoller.addWAL(dodgyWAL1);
    logRoller.addWAL(dodgyWAL2);
    // There is no 'stop' once a logRoller is running.. it just dies.
    logRoller.start();
    // Now get a region and start adding in edits.
    HTableDescriptor htd = new HTableDescriptor(TableName.META_TABLE_NAME);
    final HRegion region = initHRegion(tableName, null, null, dodgyWAL1);
    byte[] bytes = Bytes.toBytes(getName());
    NavigableMap<byte[], Integer> scopes = new TreeMap<>(Bytes.BYTES_COMPARATOR);
    scopes.put(COLUMN_FAMILY_BYTES, 0);
    MultiVersionConcurrencyControl mvcc = new MultiVersionConcurrencyControl();
    try {
        Put put = new Put(bytes);
        put.addColumn(COLUMN_FAMILY_BYTES, Bytes.toBytes("1"), bytes);
        WALKey key = new WALKey(region.getRegionInfo().getEncodedNameAsBytes(), htd.getTableName(), System.currentTimeMillis(), mvcc, scopes);
        WALEdit edit = new WALEdit();
        CellScanner CellScanner = put.cellScanner();
        assertTrue(CellScanner.advance());
        edit.add(CellScanner.current());
        LOG.info("SET throwing of exception on append");
        dodgyWAL1.throwException = true;
        // This append provokes a WAL roll request
        dodgyWAL1.append(region.getRegionInfo(), key, edit, true);
        boolean exception = false;
        try {
            dodgyWAL1.sync();
        } catch (Exception e) {
            exception = true;
        }
        assertTrue("Did not get sync exception", exception);
        // cause server abort.
        try {
            // wait LogRoller exit.
            Thread.sleep(50);
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
        final CountDownLatch latch = new CountDownLatch(1);
        // make RingBufferEventHandler sleep 1s, so the following sync
        // endOfBatch=false
        key = new WALKey(region.getRegionInfo().getEncodedNameAsBytes(), TableName.valueOf("sleep"), System.currentTimeMillis(), mvcc, scopes);
        dodgyWAL2.append(region.getRegionInfo(), key, edit, true);
        Thread t = new Thread("Sync") {

            public void run() {
                try {
                    dodgyWAL2.sync();
                } catch (IOException e) {
                    LOG.info("In sync", e);
                }
                latch.countDown();
                LOG.info("Sync exiting");
            }

            ;
        };
        t.setDaemon(true);
        t.start();
        try {
            // make sure sync have published.
            Thread.sleep(100);
        } catch (InterruptedException e1) {
            e1.printStackTrace();
        }
        // make append throw DamagedWALException
        key = new WALKey(region.getRegionInfo().getEncodedNameAsBytes(), TableName.valueOf("DamagedWALException"), System.currentTimeMillis(), mvcc, scopes);
        dodgyWAL2.append(region.getRegionInfo(), key, edit, true);
        while (latch.getCount() > 0) {
            Threads.sleep(100);
        }
        assertTrue(server.isAborted());
    } finally {
        if (logRoller != null) {
            logRoller.close();
        }
        try {
            if (region != null) {
                region.close();
            }
            if (dodgyWAL1 != null) {
                dodgyWAL1.close();
            }
            if (dodgyWAL2 != null) {
                dodgyWAL2.close();
            }
        } catch (Exception e) {
            LOG.info("On way out", e);
        }
    }
}

Example 3 with FailedLogCloseException

use of org.apache.hadoop.hbase.regionserver.wal.FailedLogCloseException in project hbase by apache.

the class AbstractWALRoller method run.

@Override
public void run() {
    while (running) {
        long now = EnvironmentEdgeManager.currentTime();
        checkLowReplication(now);
        synchronized (this) {
            if (wals.values().stream().noneMatch(rc -> rc.needsRoll(now))) {
                try {
                    wait(this.threadWakeFrequency);
                } catch (InterruptedException e) {
                    // restore the interrupt state
                    Thread.currentThread().interrupt();
                }
                // several WALs, and also check whether we should quit.
                continue;
            }
        }
        try {
            for (Iterator<Entry<WAL, RollController>> iter = wals.entrySet().iterator(); iter.hasNext(); ) {
                Entry<WAL, RollController> entry = iter.next();
                WAL wal = entry.getKey();
                RollController controller = entry.getValue();
                if (controller.isRollRequested()) {
                    // WAL roll requested, fall through
                    LOG.debug("WAL {} roll requested", wal);
                } else if (controller.needsPeriodicRoll(now)) {
                    // Time for periodic roll, fall through
                    LOG.debug("WAL {} roll period {} ms elapsed", wal, this.rollPeriod);
                } else {
                    continue;
                }
                try {
                    // Force the roll if the logroll.period is elapsed or if a roll was requested.
                    // The returned value is an collection of actual region and family names.
                    Map<byte[], List<byte[]>> regionsToFlush = controller.rollWal(now);
                    if (regionsToFlush != null) {
                        for (Map.Entry<byte[], List<byte[]>> r : regionsToFlush.entrySet()) {
                            scheduleFlush(Bytes.toString(r.getKey()), r.getValue());
                        }
                    }
                } catch (WALClosedException e) {
                    LOG.warn("WAL has been closed. Skipping rolling of writer and just remove it", e);
                    iter.remove();
                }
            }
        } catch (FailedLogCloseException | ConnectException e) {
            abort("Failed log close in log roller", e);
        } catch (IOException ex) {
            // Abort if we get here. We probably won't recover an IOE. HBASE-1132
            abort("IOE in log roller", ex instanceof RemoteException ? ((RemoteException) ex).unwrapRemoteException() : ex);
        } catch (Exception ex) {
            LOG.error("Log rolling failed", ex);
            abort("Log rolling failed", ex);
        }
    }
    LOG.info("LogRoller exiting.");
}