Examples with BookieSocketAddress org.apache.bookkeeper.net.BookieSocketAddress used on opensource projects

Example 51 with BookieSocketAddress

use of org.apache.bookkeeper.net.BookieSocketAddress in project bookkeeper by apache.

the class BookieRecoveryTest method testBookieRecoveryOnOpenedLedgers.

@Test
public void testBookieRecoveryOnOpenedLedgers() throws Exception {
    // Create the ledgers
    int numLedgers = 3;
    List<LedgerHandle> lhs = createLedgers(numLedgers, numBookies, 2);
    // Write the entries for the ledgers with dummy values
    int numMsgs = 10;
    writeEntriestoLedgers(numMsgs, 0, lhs);
    // Shutdown the first bookie server
    ArrayList<BookieSocketAddress> lastEnsemble = lhs.get(0).getLedgerMetadata().getEnsembles().entrySet().iterator().next().getValue();
    BookieSocketAddress bookieToKill = lastEnsemble.get(lastEnsemble.size() - 1);
    killBookie(bookieToKill);
    // start a new bookie
    startNewBookie();
    LOG.info("Now recover the data on the killed bookie (" + bookieToKill + ") and replicate it to a random available one");
    bkAdmin.recoverBookieData(bookieToKill);
    for (LedgerHandle lh : lhs) {
        assertTrue("Not fully replicated", verifyFullyReplicated(lh, numMsgs));
    }
    try {
        // we can't write entries
        writeEntriestoLedgers(numMsgs, 0, lhs);
        fail("should not reach here");
    } catch (Exception e) {
    }
}

Example 52 with BookieSocketAddress

use of org.apache.bookkeeper.net.BookieSocketAddress in project bookkeeper by apache.

the class LedgerRecoveryTest method testLedgerRecoveryWithRollingRestart.

/**
 * {@link https://issues.apache.org/jira/browse/BOOKKEEPER-355}
 * A recovery during a rolling restart shouldn't affect the ability
 * to recovery the ledger later.
 * We have a ledger on ensemble B1,B2,B3.
 * The sequence of events is
 * 1. B1 brought down for maintenance
 * 2. Ledger recovery started
 * 3. B2 answers read last confirmed.
 * 4. B1 replaced in ensemble by B4
 * 5. Write to B4 fails for some reason
 * 6. B1 comes back up.
 * 7. B2 goes down for maintenance.
 * 8. Ledger recovery starts (ledger is now unavailable)
 */
@Test
public void testLedgerRecoveryWithRollingRestart() throws Exception {
    LedgerHandle lhbefore = bkc.createLedger(numBookies, 2, digestType, "".getBytes());
    for (int i = 0; i < (numBookies * 3) + 1; i++) {
        lhbefore.addEntry("data".getBytes());
    }
    // Add a dead bookie to the cluster
    ServerConfiguration conf = newServerConfiguration();
    Bookie deadBookie1 = new Bookie(conf) {

        @Override
        public void recoveryAddEntry(ByteBuf entry, WriteCallback cb, Object ctx, byte[] masterKey) throws IOException, BookieException {
            // drop request to simulate a slow and failed bookie
            throw new IOException("Couldn't write for some reason");
        }
    };
    bsConfs.add(conf);
    bs.add(startBookie(conf, deadBookie1));
    // kill first bookie server
    BookieSocketAddress bookie1 = lhbefore.getLedgerMetadata().currentEnsemble.get(0);
    ServerConfiguration conf1 = killBookie(bookie1);
    // ensemble in the ensemble, and another bookie is available in zk, but not writtable
    try {
        bkc.openLedger(lhbefore.getId(), digestType, "".getBytes());
        fail("Shouldn't be able to open ledger, there should be entries missing");
    } catch (BKException.BKLedgerRecoveryException e) {
    // expected
    }
    // restart the first server, kill the second
    bsConfs.add(conf1);
    bs.add(startBookie(conf1));
    BookieSocketAddress bookie2 = lhbefore.getLedgerMetadata().currentEnsemble.get(1);
    ServerConfiguration conf2 = killBookie(bookie2);
    // using async, because this could trigger an assertion
    final AtomicInteger returnCode = new AtomicInteger(0);
    final CountDownLatch openLatch = new CountDownLatch(1);
    bkc.asyncOpenLedger(lhbefore.getId(), digestType, "".getBytes(), new AsyncCallback.OpenCallback() {

        public void openComplete(int rc, LedgerHandle lh, Object ctx) {
            returnCode.set(rc);
            openLatch.countDown();
            if (rc == BKException.Code.OK) {
                try {
                    lh.close();
                } catch (Exception e) {
                    LOG.error("Exception closing ledger handle", e);
                }
            }
        }
    }, null);
    assertTrue("Open call should have completed", openLatch.await(5, TimeUnit.SECONDS));
    assertFalse("Open should not have succeeded", returnCode.get() == BKException.Code.OK);
    bsConfs.add(conf2);
    bs.add(startBookie(conf2));
    LedgerHandle lhafter = bkc.openLedger(lhbefore.getId(), digestType, "".getBytes());
    assertEquals("Fenced ledger should have correct lastAddConfirmed", lhbefore.getLastAddConfirmed(), lhafter.getLastAddConfirmed());
}

Example 53 with BookieSocketAddress

use of org.apache.bookkeeper.net.BookieSocketAddress in project bookkeeper by apache.

the class TestDelayEnsembleChange method verifyEntriesRange.

private void verifyEntriesRange(LedgerHandle lh, long startEntry, long untilEntry, long expectedSuccess, long expectedMissing) throws Exception {
    LedgerMetadata md = lh.getLedgerMetadata();
    for (long eid = startEntry; eid < untilEntry; eid++) {
        ArrayList<BookieSocketAddress> addresses = md.getEnsemble(eid);
        VerificationCallback callback = new VerificationCallback(addresses.size());
        for (BookieSocketAddress addr : addresses) {
            bkc.getBookieClient().readEntry(addr, lh.getId(), eid, callback, addr, 0, null);
        }
        callback.latch.await();
        assertTrue(expectedSuccess >= callback.numSuccess.get());
        assertTrue(expectedMissing <= callback.numMissing.get());
        assertEquals(0, callback.numFailure.get());
    }
}

Example 54 with BookieSocketAddress

use of org.apache.bookkeeper.net.BookieSocketAddress in project bookkeeper by apache.

the class TestDelayEnsembleChange method testChangeEnsembleSecondBookieReadOnly.

@Test
public void testChangeEnsembleSecondBookieReadOnly() throws Exception {
    LedgerHandle lh = bkc.createLedger(3, 3, 2, digestType, testPasswd);
    byte[] data = "foobar".getBytes();
    int numEntries = 10;
    for (int i = 0; i < numEntries; i++) {
        lh.addEntry(data);
    }
    BookieSocketAddress failedBookie = lh.getLedgerMetadata().currentEnsemble.get(0);
    BookieSocketAddress readOnlyBookie = lh.getLedgerMetadata().currentEnsemble.get(1);
    ServerConfiguration conf0 = killBookie(failedBookie);
    for (int i = 0; i < numEntries; i++) {
        lh.addEntry(data);
    }
    assertEquals("There should be ensemble change if delaying ensemble change is enabled.", 1, lh.getLedgerMetadata().getEnsembles().size());
    // kill two bookies, but we still have 3 bookies for the ack quorum.
    setBookieToReadOnly(readOnlyBookie);
    for (int i = 0; i < numEntries; i++) {
        lh.addEntry(data);
    }
    // ensure there is no ensemble changed
    assertEquals("The ensemble should change when a bookie is readonly even if we delay ensemble change.", 2, lh.getLedgerMetadata().getEnsembles().size());
    assertEquals(3, lh.getLedgerMetadata().currentEnsemble.size());
    assertFalse(lh.getLedgerMetadata().currentEnsemble.contains(failedBookie));
    assertFalse(lh.getLedgerMetadata().currentEnsemble.contains(readOnlyBookie));
}

Example 55 with BookieSocketAddress

use of org.apache.bookkeeper.net.BookieSocketAddress in project bookkeeper by apache.

the class TestDelayEnsembleChange method testChangeEnsembleIfBrokenAckQuorum.

@Test
public void testChangeEnsembleIfBrokenAckQuorum() throws Exception {
    startNewBookie();
    startNewBookie();
    startNewBookie();
    bkc.getTestStatsProvider().clear();
    LedgerHandle lh = bkc.createLedger(5, 5, 3, digestType, testPasswd);
    byte[] data = "foobar".getBytes();
    int numEntries = 5;
    for (int i = 0; i < numEntries; i++) {
        lh.addEntry(data);
    }
    for (BookieSocketAddress addr : lh.getLedgerMetadata().getEnsembles().get(0L)) {
        assertTrue(LEDGER_ENSEMBLE_BOOKIE_DISTRIBUTION + " should be > 0 for " + addr, bkc.getTestStatsProvider().getCounter(CLIENT_SCOPE + "." + LEDGER_ENSEMBLE_BOOKIE_DISTRIBUTION + "-" + addr).get() > 0);
    }
    assertTrue("Stats should have captured a new ensemble", bkc.getTestStatsProvider().getOpStatsLogger(CLIENT_SCOPE + "." + WATCHER_SCOPE + "." + NEW_ENSEMBLE_TIME).getSuccessCount() > 0);
    assertTrue("Stats should not have captured an ensemble change", bkc.getTestStatsProvider().getOpStatsLogger(CLIENT_SCOPE + "." + WATCHER_SCOPE + "." + REPLACE_BOOKIE_TIME).getSuccessCount() == 0);
    logger.info("Kill bookie 0 and write {} entries.", numEntries);
    // kill two bookies, but we still have 3 bookies for the ack quorum.
    ServerConfiguration conf0 = killBookie(lh.getLedgerMetadata().currentEnsemble.get(0));
    for (int i = numEntries; i < 2 * numEntries; i++) {
        lh.addEntry(data);
    }
    // ensure there is no ensemble changed
    assertEquals("There should be no ensemble change if delaying ensemble change is enabled.", 1, lh.getLedgerMetadata().getEnsembles().size());
    assertTrue("Stats should not have captured an ensemble change", bkc.getTestStatsProvider().getOpStatsLogger(CLIENT_SCOPE + "." + WATCHER_SCOPE + "." + REPLACE_BOOKIE_TIME).getSuccessCount() == 0);
    logger.info("Kill bookie 1 and write another {} entries.", numEntries);
    ServerConfiguration conf1 = killBookie(lh.getLedgerMetadata().currentEnsemble.get(1));
    for (int i = 2 * numEntries; i < 3 * numEntries; i++) {
        lh.addEntry(data);
    }
    // ensure there is no ensemble changed
    assertEquals("There should be no ensemble change if delaying ensemble change is enabled.", 1, lh.getLedgerMetadata().getEnsembles().size());
    logger.info("Kill bookie 2 and write another {} entries.", numEntries);
    ServerConfiguration conf2 = killBookie(lh.getLedgerMetadata().currentEnsemble.get(2));
    for (int i = 3 * numEntries; i < 4 * numEntries; i++) {
        lh.addEntry(data);
    }
    // ensemble change should kick in
    assertEquals("There should be ensemble change if ack quorum couldn't be formed.", 2, lh.getLedgerMetadata().getEnsembles().size());
    assertTrue("Stats should have captured an ensemble change", bkc.getTestStatsProvider().getOpStatsLogger(CLIENT_SCOPE + "." + WATCHER_SCOPE + "." + REPLACE_BOOKIE_TIME).getSuccessCount() > 0);
    ArrayList<BookieSocketAddress> firstFragment = lh.getLedgerMetadata().getEnsemble(0);
    ArrayList<BookieSocketAddress> secondFragment = lh.getLedgerMetadata().getEnsemble(3 * numEntries);
    assertFalse(firstFragment.get(0).equals(secondFragment.get(0)));
    assertFalse(firstFragment.get(1).equals(secondFragment.get(1)));
    assertFalse(firstFragment.get(2).equals(secondFragment.get(2)));
    assertEquals(firstFragment.get(3), secondFragment.get(3));
    assertEquals(firstFragment.get(4), secondFragment.get(4));
    bsConfs.add(conf0);
    bs.add(startBookie(conf0));
    bsConfs.add(conf1);
    bs.add(startBookie(conf1));
    bsConfs.add(conf2);
    bs.add(startBookie(conf2));
    for (int i = 4 * numEntries; i < 5 * numEntries; i++) {
        lh.addEntry(data);
    }
    // ensure there is no ensemble changed
    assertEquals("There should be no ensemble change if delaying ensemble change is enabled.", 2, lh.getLedgerMetadata().getEnsembles().size());
    // check entries
    verifyEntries(lh, 0, numEntries, 5, 0);
    verifyEntries(lh, numEntries, 2 * numEntries, 4, 1);
    verifyEntries(lh, 2 * numEntries, 3 * numEntries, 3, 2);
    verifyEntries(lh, 3 * numEntries, 4 * numEntries, 5, 0);
    verifyEntries(lh, 4 * numEntries, 5 * numEntries, 5, 0);
}