Examples with UnavailableException org.apache.bookkeeper.replication.ReplicationException.UnavailableException used on opensource projects

Example 11 with UnavailableException

use of org.apache.bookkeeper.replication.ReplicationException.UnavailableException in project bookkeeper by apache.

the class BookKeeperClusterTestCase method startBookie.

/**
 * Start a bookie with the given bookie instance. Also, starts the auto
 * recovery for this bookie, if isAutoRecoveryEnabled is true.
 */
protected BookieServer startBookie(ServerConfiguration conf, final Bookie b) throws Exception {
    TestStatsProvider provider = new TestStatsProvider();
    BookieServer server = new BookieServer(conf, provider.getStatsLogger("")) {

        @Override
        protected Bookie newBookie(ServerConfiguration conf) {
            return b;
        }
    };
    BookieSocketAddress address = Bookie.getBookieAddress(conf);
    if (bkc == null) {
        bkc = new BookKeeperTestClient(baseClientConf, new TestStatsProvider());
    }
    Future<?> waitForBookie = conf.isForceReadOnlyBookie() ? bkc.waitForReadOnlyBookie(address) : bkc.waitForWritableBookie(address);
    server.start();
    bsLoggers.put(server.getLocalAddress(), provider);
    waitForBookie.get(30, TimeUnit.SECONDS);
    LOG.info("New bookie '{}' has been created.", address);
    try {
        startAutoRecovery(server, conf);
    } catch (CompatibilityException ce) {
        LOG.error("Exception while starting AutoRecovery!", ce);
    } catch (UnavailableException ue) {
        LOG.error("Exception while starting AutoRecovery!", ue);
    }
    return server;
}

Example 12 with UnavailableException

use of org.apache.bookkeeper.replication.ReplicationException.UnavailableException in project bookkeeper by apache.

the class AuditorElector method submitElectionTask.

/**
 * Performing the auditor election using the ZooKeeper ephemeral sequential
 * znode. The bookie which has created the least sequential will be elect as
 * Auditor.
 */
@VisibleForTesting
void submitElectionTask() {
    Runnable r = new Runnable() {

        public void run() {
            if (!running.get()) {
                return;
            }
            try {
                // creating my vote in zk. Vote format is 'V_numeric'
                createMyVote();
                List<String> children = zkc.getChildren(getVotePath(""), false);
                if (0 >= children.size()) {
                    throw new IllegalArgumentException("Atleast one bookie server should present to elect the Auditor!");
                }
                // sorting in ascending order of sequential number
                Collections.sort(children, new ElectionComparator());
                String voteNode = StringUtils.substringAfterLast(myVote, PATH_SEPARATOR);
                // starting Auditing service
                if (children.get(AUDITOR_INDEX).equals(voteNode)) {
                    // update the auditor bookie id in the election path. This is
                    // done for debugging purpose
                    AuditorVoteFormat.Builder builder = AuditorVoteFormat.newBuilder().setBookieId(bookieId);
                    zkc.setData(getVotePath(""), TextFormat.printToString(builder.build()).getBytes(UTF_8), -1);
                    auditor = new Auditor(bookieId, conf, zkc, statsLogger);
                    auditor.start();
                } else {
                    // If not an auditor, will be watching to my predecessor and
                    // looking the previous node deletion.
                    Watcher electionWatcher = new ElectionWatcher();
                    int myIndex = children.indexOf(voteNode);
                    int prevNodeIndex = myIndex - 1;
                    if (null == zkc.exists(getVotePath(PATH_SEPARATOR) + children.get(prevNodeIndex), electionWatcher)) {
                        // While adding, the previous znode doesn't exists.
                        // Again going to election.
                        submitElectionTask();
                    }
                    electionAttempts.inc();
                }
            } catch (KeeperException e) {
                LOG.error("Exception while performing auditor election", e);
                submitShutdownTask();
            } catch (InterruptedException e) {
                LOG.error("Interrupted while performing auditor election", e);
                Thread.currentThread().interrupt();
                submitShutdownTask();
            } catch (UnavailableException e) {
                LOG.error("Ledger underreplication manager unavailable during election", e);
                submitShutdownTask();
            }
        }
    };
    executor.submit(r);
}

Example 13 with UnavailableException

use of org.apache.bookkeeper.replication.ReplicationException.UnavailableException in project bookkeeper by apache.

the class Auditor method submitAuditTask.

@VisibleForTesting
synchronized Future<?> submitAuditTask() {
    if (executor.isShutdown()) {
        SettableFuture<Void> f = SettableFuture.<Void>create();
        f.setException(new BKAuditException("Auditor shutting down"));
        return f;
    }
    return executor.submit(new Runnable() {

        @SuppressWarnings("unchecked")
        public void run() {
            try {
                waitIfLedgerReplicationDisabled();
                int lostBookieRecoveryDelay = Auditor.this.ledgerUnderreplicationManager.getLostBookieRecoveryDelay();
                List<String> availableBookies = getAvailableBookies();
                // casting to String, as knownBookies and availableBookies
                // contains only String values
                // find new bookies(if any) and update the known bookie list
                Collection<String> newBookies = CollectionUtils.subtract(availableBookies, knownBookies);
                knownBookies.addAll(newBookies);
                if (!bookiesToBeAudited.isEmpty() && knownBookies.containsAll(bookiesToBeAudited)) {
                    // check should be ok
                    if (auditTask != null && auditTask.cancel(false)) {
                        auditTask = null;
                        numDelayedBookieAuditsCancelled.inc();
                    }
                    bookiesToBeAudited.clear();
                }
                // find lost bookies(if any)
                bookiesToBeAudited.addAll(CollectionUtils.subtract(knownBookies, availableBookies));
                if (bookiesToBeAudited.size() == 0) {
                    return;
                }
                knownBookies.removeAll(bookiesToBeAudited);
                if (lostBookieRecoveryDelay == 0) {
                    startAudit(false);
                    bookiesToBeAudited.clear();
                    return;
                }
                if (bookiesToBeAudited.size() > 1) {
                    // if more than one bookie is down, start the audit immediately;
                    LOG.info("Multiple bookie failure; not delaying bookie audit. " + "Bookies lost now: {}; All lost bookies: {}", CollectionUtils.subtract(knownBookies, availableBookies), bookiesToBeAudited);
                    if (auditTask != null && auditTask.cancel(false)) {
                        auditTask = null;
                        numDelayedBookieAuditsCancelled.inc();
                    }
                    startAudit(false);
                    bookiesToBeAudited.clear();
                    return;
                }
                if (auditTask == null) {
                    // if there is no scheduled audit, schedule one
                    auditTask = executor.schedule(new Runnable() {

                        public void run() {
                            startAudit(false);
                            auditTask = null;
                            bookiesToBeAudited.clear();
                        }
                    }, lostBookieRecoveryDelay, TimeUnit.SECONDS);
                    numBookieAuditsDelayed.inc();
                    LOG.info("Delaying bookie audit by {} secs for {}", lostBookieRecoveryDelay, bookiesToBeAudited);
                }
            } catch (BKException bke) {
                LOG.error("Exception getting bookie list", bke);
            } catch (InterruptedException ie) {
                Thread.currentThread().interrupt();
                LOG.error("Interrupted while watching available bookies ", ie);
            } catch (UnavailableException ue) {
                LOG.error("Exception while watching available bookies", ue);
            }
        }
    });
}

Example 14 with UnavailableException

use of org.apache.bookkeeper.replication.ReplicationException.UnavailableException in project bookkeeper by apache.

the class BookKeeperAdminTest method testTriggerAudit.

@Test
public void testTriggerAudit() throws Exception {
    ZkLedgerUnderreplicationManager urLedgerMgr = new ZkLedgerUnderreplicationManager(baseClientConf, zkc);
    BookKeeperAdmin bkAdmin = new BookKeeperAdmin(zkUtil.getZooKeeperConnectString());
    int lostBookieRecoveryDelayValue = bkAdmin.getLostBookieRecoveryDelay();
    urLedgerMgr.disableLedgerReplication();
    try {
        bkAdmin.triggerAudit();
        fail("Trigger Audit should have failed because LedgerReplication is disabled");
    } catch (UnavailableException une) {
    // expected
    }
    assertEquals("LostBookieRecoveryDelay", lostBookieRecoveryDelayValue, bkAdmin.getLostBookieRecoveryDelay());
    urLedgerMgr.enableLedgerReplication();
    bkAdmin.triggerAudit();
    assertEquals("LostBookieRecoveryDelay", lostBookieRecoveryDelayValue, bkAdmin.getLostBookieRecoveryDelay());
    long ledgerId = 1L;
    LedgerHandle ledgerHandle = bkc.createLedgerAdv(ledgerId, numBookies, numBookies, numBookies, digestType, PASSWORD.getBytes(), null);
    ledgerHandle.addEntry(0, "data".getBytes());
    ledgerHandle.close();
    killBookie(1);
    /*
         * since lostBookieRecoveryDelay is set, when a bookie is died, it will
         * not start Audit process immediately. But when triggerAudit is called
         * it will force audit process.
         */
    bkAdmin.triggerAudit();
    Thread.sleep(500);
    Iterator<Long> ledgersToRereplicate = urLedgerMgr.listLedgersToRereplicate(null);
    assertTrue("There are supposed to be underreplicatedledgers", ledgersToRereplicate.hasNext());
    assertEquals("Underreplicated ledgerId", ledgerId, ledgersToRereplicate.next().longValue());
    bkAdmin.close();
}

Example 15 with UnavailableException

use of org.apache.bookkeeper.replication.ReplicationException.UnavailableException in project bookkeeper by apache.

the class BookKeeperAdmin method triggerAudit.

/**
 * Trigger AuditTask by resetting lostBookieRecoveryDelay to its current
 * value. If Autorecovery is not enabled or if there is no Auditor then this
 * method will throw UnavailableException.
 *
 * @throws CompatibilityException
 * @throws KeeperException
 * @throws InterruptedException
 * @throws UnavailableException
 * @throws IOException
 */
public void triggerAudit() throws CompatibilityException, KeeperException, InterruptedException, UnavailableException, IOException {
    LedgerUnderreplicationManager urlManager = getUnderreplicationManager();
    if (!urlManager.isLedgerReplicationEnabled()) {
        LOG.error("Autorecovery is disabled. So giving up!");
        throw new UnavailableException("Autorecovery is disabled. So giving up!");
    }
    BookieSocketAddress auditorId = AuditorElector.getCurrentAuditor(new ServerConfiguration(bkc.conf), bkc.getZkHandle());
    if (auditorId == null) {
        LOG.error("No auditor elected, though Autorecovery is enabled. So giving up.");
        throw new UnavailableException("No auditor elected, though Autorecovery is enabled. So giving up.");
    }
    int previousLostBookieRecoveryDelayValue = urlManager.getLostBookieRecoveryDelay();
    LOG.info("Resetting LostBookieRecoveryDelay value: {}, to kickstart audit task", previousLostBookieRecoveryDelayValue);
    urlManager.setLostBookieRecoveryDelay(previousLostBookieRecoveryDelayValue);
}