Examples with JournalAndStream org.apache.hadoop.hdfs.server.namenode.JournalSet.JournalAndStream used on opensource projects

Example 1 with JournalAndStream

use of org.apache.hadoop.hdfs.server.namenode.JournalSet.JournalAndStream in project hadoop by apache.

the class TestEditLogJournalFailures method testSingleRequiredFailedEditsDirOnSetReadyToFlush.

@Test
public void testSingleRequiredFailedEditsDirOnSetReadyToFlush() throws IOException {
    // Set one of the edits dirs to be required.
    String[] editsDirs = cluster.getConfiguration(0).getTrimmedStrings(DFSConfigKeys.DFS_NAMENODE_NAME_DIR_KEY);
    shutDownMiniCluster();
    Configuration conf = getConf();
    conf.set(DFSConfigKeys.DFS_NAMENODE_EDITS_DIR_REQUIRED_KEY, editsDirs[0]);
    conf.setInt(DFSConfigKeys.DFS_NAMENODE_EDITS_DIR_MINIMUM_KEY, 0);
    conf.setInt(DFSConfigKeys.DFS_NAMENODE_CHECKED_VOLUMES_MINIMUM_KEY, 0);
    setUpMiniCluster(conf, true);
    assertTrue(doAnEdit());
    // Invalidated the one required edits journal.
    invalidateEditsDirAtIndex(0, false, false);
    JournalAndStream nonRequiredJas = getJournalAndStream(1);
    EditLogFileOutputStream nonRequiredSpy = spyOnStream(nonRequiredJas);
    // The NN has not terminated (no ExitException thrown)
    // ..and that the other stream is active.
    assertTrue(nonRequiredJas.isActive());
    try {
        doAnEdit();
        fail("A single failure of a required journal should have halted the NN");
    } catch (RemoteException re) {
        assertTrue(re.getClassName().contains("ExitException"));
        GenericTestUtils.assertExceptionContains("setReadyToFlush failed for required journal", re);
    }
    // Since the required directory failed setReadyToFlush, and that
    // directory was listed prior to the non-required directory,
    // we should not call setReadyToFlush on the non-required
    // directory. Regression test for HDFS-2874.
    Mockito.verify(nonRequiredSpy, Mockito.never()).setReadyToFlush();
    assertFalse(nonRequiredJas.isActive());
}

Example 2 with JournalAndStream

use of org.apache.hadoop.hdfs.server.namenode.JournalSet.JournalAndStream in project hadoop by apache.

the class TestEditLogJournalFailures method invalidateEditsDirAtIndex.

/**
   * Replace the journal at index <code>index</code> with one that throws an
   * exception on flush.
   * 
   * @param index the index of the journal to take offline.
   * @return the original <code>EditLogOutputStream</code> of the journal.
   */
private void invalidateEditsDirAtIndex(int index, boolean failOnFlush, boolean failOnWrite) throws IOException {
    JournalAndStream jas = getJournalAndStream(index);
    EditLogFileOutputStream spyElos = spyOnStream(jas);
    if (failOnWrite) {
        doThrow(new IOException("fail on write()")).when(spyElos).write((FSEditLogOp) any());
    }
    if (failOnFlush) {
        doThrow(new IOException("fail on flush()")).when(spyElos).flush();
    } else {
        doThrow(new IOException("fail on setReadyToFlush()")).when(spyElos).setReadyToFlush();
    }
}

Example 3 with JournalAndStream

use of org.apache.hadoop.hdfs.server.namenode.JournalSet.JournalAndStream in project hadoop by apache.

the class TestEditLogRace method testSaveImageWhileSyncInProgress.

/**
   * The logSync() method in FSEditLog is unsynchronized whiel syncing
   * so that other threads can concurrently enqueue edits while the prior
   * sync is ongoing. This test checks that the log is saved correctly
   * if the saveImage occurs while the syncing thread is in the unsynchronized middle section.
   * 
   * This replicates the following manual test proposed by Konstantin:
   *   I start the name-node in debugger.
   *   I do -mkdir and stop the debugger in logSync() just before it does flush.
   *   Then I enter safe mode with another client
   *   I start saveNamepsace and stop the debugger in
   *     FSImage.saveFSImage() -> FSEditLog.createEditLogFile()
   *     -> EditLogFileOutputStream.create() ->
   *     after truncating the file but before writing LAYOUT_VERSION into it.
   *   Then I let logSync() run.
   *   Then I terminate the name-node.
   *   After that the name-node wont start, since the edits file is broken.
   */
@Test
public void testSaveImageWhileSyncInProgress() throws Exception {
    Configuration conf = getConf();
    NameNode.initMetrics(conf, NamenodeRole.NAMENODE);
    DFSTestUtil.formatNameNode(conf);
    final FSNamesystem namesystem = FSNamesystem.loadFromDisk(conf);
    try {
        FSImage fsimage = namesystem.getFSImage();
        FSEditLog editLog = fsimage.getEditLog();
        JournalAndStream jas = editLog.getJournals().get(0);
        EditLogFileOutputStream spyElos = spy((EditLogFileOutputStream) jas.getCurrentStream());
        jas.setCurrentStreamForTests(spyElos);
        final AtomicReference<Throwable> deferredException = new AtomicReference<Throwable>();
        final CountDownLatch waitToEnterFlush = new CountDownLatch(1);
        final Thread doAnEditThread = new Thread() {

            @Override
            public void run() {
                try {
                    LOG.info("Starting mkdirs");
                    namesystem.mkdirs("/test", new PermissionStatus("test", "test", new FsPermission((short) 00755)), true);
                    LOG.info("mkdirs complete");
                } catch (Throwable ioe) {
                    LOG.fatal("Got exception", ioe);
                    deferredException.set(ioe);
                    waitToEnterFlush.countDown();
                }
            }
        };
        Answer<Void> blockingFlush = new Answer<Void>() {

            @Override
            public Void answer(InvocationOnMock invocation) throws Throwable {
                LOG.info("Flush called");
                if (useAsyncEditLog || Thread.currentThread() == doAnEditThread) {
                    LOG.info("edit thread: Telling main thread we made it to flush section...");
                    // Signal to main thread that the edit thread is in the racy section
                    waitToEnterFlush.countDown();
                    LOG.info("edit thread: sleeping for " + BLOCK_TIME + "secs");
                    Thread.sleep(BLOCK_TIME * 1000);
                    LOG.info("Going through to flush. This will allow the main thread to continue.");
                }
                invocation.callRealMethod();
                LOG.info("Flush complete");
                return null;
            }
        };
        doAnswer(blockingFlush).when(spyElos).flush();
        doAnEditThread.start();
        // Wait for the edit thread to get to the logsync unsynchronized section
        LOG.info("Main thread: waiting to enter flush...");
        waitToEnterFlush.await();
        assertNull(deferredException.get());
        LOG.info("Main thread: detected that logSync is in unsynchronized section.");
        LOG.info("Trying to enter safe mode.");
        LOG.info("This should block for " + BLOCK_TIME + "sec, since flush will sleep that long");
        long st = Time.now();
        namesystem.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
        long et = Time.now();
        LOG.info("Entered safe mode");
        // Make sure we really waited for the flush to complete!
        assertTrue(et - st > (BLOCK_TIME - 1) * 1000);
        // Once we're in safe mode, save namespace.
        namesystem.saveNamespace(0, 0);
        LOG.info("Joining on edit thread...");
        doAnEditThread.join();
        assertNull(deferredException.get());
        // We did 3 edits: begin, txn, and end
        assertEquals(3, verifyEditLogs(namesystem, fsimage, NNStorage.getFinalizedEditsFileName(1, 3), 1));
        // after the save, just the one "begin"
        assertEquals(1, verifyEditLogs(namesystem, fsimage, NNStorage.getInProgressEditsFileName(4), 4));
    } finally {
        LOG.info("Closing nn");
        if (namesystem != null)
            namesystem.close();
    }
}

Example 4 with JournalAndStream

use of org.apache.hadoop.hdfs.server.namenode.JournalSet.JournalAndStream in project hadoop by apache.

the class TestStorageRestore method invalidateStorage.

/**
   * invalidate storage by removing the second and third storage directories
   */
public void invalidateStorage(FSImage fi, Set<File> filesToInvalidate) throws IOException {
    ArrayList<StorageDirectory> al = new ArrayList<StorageDirectory>(2);
    Iterator<StorageDirectory> it = fi.getStorage().dirIterator();
    while (it.hasNext()) {
        StorageDirectory sd = it.next();
        if (filesToInvalidate.contains(sd.getRoot())) {
            LOG.info("causing IO error on " + sd.getRoot());
            al.add(sd);
        }
    }
    // simulate an error
    fi.getStorage().reportErrorsOnDirectories(al);
    for (JournalAndStream j : fi.getEditLog().getJournals()) {
        if (j.getManager() instanceof FileJournalManager) {
            FileJournalManager fm = (FileJournalManager) j.getManager();
            if (fm.getStorageDirectory().getRoot().equals(path2) || fm.getStorageDirectory().getRoot().equals(path3)) {
                EditLogOutputStream mockStream = spy(j.getCurrentStream());
                j.setCurrentStreamForTests(mockStream);
                doThrow(new IOException("Injected fault: write")).when(mockStream).write(Mockito.<FSEditLogOp>anyObject());
            }
        }
    }
}

Example 5 with JournalAndStream

use of org.apache.hadoop.hdfs.server.namenode.JournalSet.JournalAndStream in project hadoop by apache.

the class FSNamesystem method getNameJournalStatus.

// NameNodeMXBean
@Override
public String getNameJournalStatus() {
    List<Map<String, String>> jasList = new ArrayList<Map<String, String>>();
    FSEditLog log = getFSImage().getEditLog();
    if (log != null) {
        // This flag can be false because we cannot hold a lock of FSEditLog
        // for metrics.
        boolean openForWrite = log.isOpenForWriteWithoutLock();
        for (JournalAndStream jas : log.getJournals()) {
            final Map<String, String> jasMap = new HashMap<String, String>();
            String manager = jas.getManager().toString();
            jasMap.put("required", String.valueOf(jas.isRequired()));
            jasMap.put("disabled", String.valueOf(jas.isDisabled()));
            jasMap.put("manager", manager);
            if (jas.isDisabled()) {
                jasMap.put("stream", "Failed");
            } else if (openForWrite) {
                EditLogOutputStream elos = jas.getCurrentStream();
                if (elos != null) {
                    jasMap.put("stream", elos.generateReport());
                } else {
                    jasMap.put("stream", "not currently writing");
                }
            } else {
                jasMap.put("stream", "open for read");
            }
            jasList.add(jasMap);
        }
    }
    return JSON.toString(jasList);
}