Examples with BookkeeperCommitLog herddb.cluster.BookkeeperCommitLog used on opensource projects

Example 6 with BookkeeperCommitLog

use of herddb.cluster.BookkeeperCommitLog in project herddb by diennea.

the class BookKeeperCommitLogTest method testFollowEmptyLedgerBookieDown.

@Test
public void testFollowEmptyLedgerBookieDown() throws Exception {
    BookieId secondBookie = testEnv.startNewBookie();
    final String tableSpaceUUID = UUID.randomUUID().toString();
    final String name = TableSpace.DEFAULT;
    final String nodeid = "nodeid";
    ServerConfiguration serverConfiguration = newServerConfigurationWithAutoPort();
    try (ZookeeperMetadataStorageManager man = new ZookeeperMetadataStorageManager(testEnv.getAddress(), testEnv.getTimeout(), testEnv.getPath());
        BookkeeperCommitLogManager logManager = new BookkeeperCommitLogManager(man, serverConfiguration, NullStatsLogger.INSTANCE)) {
        logManager.setEnsemble(2);
        logManager.setWriteQuorumSize(2);
        logManager.setAckQuorumSize(2);
        man.start();
        logManager.start();
        try (BookkeeperCommitLog writer = logManager.createCommitLog(tableSpaceUUID, name, nodeid)) {
            writer.startWriting(1);
            // create a ledger, up to 0.14.x no "logical" write happens, so Bookies are not aware of the
            // the ledger
            // stop one bookie
            testEnv.stopBookie(secondBookie);
            // but since 0.15.0 we are now writing a NOOP entry at the beginning of the ledger in order to workaround this issue.
            try (BookkeeperCommitLog reader = logManager.createCommitLog(tableSpaceUUID, name, nodeid)) {
                reader.recovery(LogSequenceNumber.START_OF_TIME, (a, b) -> {
                }, false);
            }
        }
    }
}

Example 7 with BookkeeperCommitLog

use of herddb.cluster.BookkeeperCommitLog in project herddb by diennea.

the class BookKeeperCommitLogTest method testBookieFailureSyncWrites.

@Test
public void testBookieFailureSyncWrites() throws Exception {
    final String tableSpaceUUID = UUID.randomUUID().toString();
    final String name = TableSpace.DEFAULT;
    final String nodeid = "nodeid";
    ServerConfiguration serverConfiguration = newServerConfigurationWithAutoPort();
    try (ZookeeperMetadataStorageManager man = new ZookeeperMetadataStorageManager(testEnv.getAddress(), testEnv.getTimeout(), testEnv.getPath());
        BookkeeperCommitLogManager logManager = new BookkeeperCommitLogManager(man, serverConfiguration, NullStatsLogger.INSTANCE)) {
        man.start();
        logManager.start();
        try (BookkeeperCommitLog writer = logManager.createCommitLog(tableSpaceUUID, name, nodeid)) {
            writer.startWriting(1);
            writer.log(LogEntryFactory.beginTransaction(1), true).getLogSequenceNumber();
            this.testEnv.pauseBookie();
            // this is deemed to fail and we will close the log
            TestUtils.assertThrows(LogNotAvailableException.class, () -> writer.log(LogEntryFactory.beginTransaction(2), true).getLogSequenceNumber());
            assertTrue(writer.isFailed());
            // this one will fail as well
            TestUtils.assertThrows(LogNotAvailableException.class, () -> writer.log(LogEntryFactory.beginTransaction(2), true).getLogSequenceNumber());
            // no way to recover this instance of BookkeeperCommitLog
            // we will bounce the leader and it will restart with a full recovery
            // in a production env you do not have only one bookie, and the failure
            // of a single bookie will be handled with an ensemble change
            this.testEnv.resumeBookie();
            TestUtils.assertThrows(LogNotAvailableException.class, () -> writer.log(LogEntryFactory.beginTransaction(2), true).getLogSequenceNumber());
        }
        // check expected reads
        try (BookkeeperCommitLog reader = logManager.createCommitLog(tableSpaceUUID, name, nodeid)) {
            List<Map.Entry<LogSequenceNumber, LogEntry>> list = new ArrayList<>();
            reader.recovery(LogSequenceNumber.START_OF_TIME, (lsn, entry) -> {
                if (entry.type != LogEntryType.NOOP) {
                    list.add(new AbstractMap.SimpleImmutableEntry<>(lsn, entry));
                }
            }, false);
            assertEquals(1, list.size());
            assertTrue(list.get(0).getKey().after(LogSequenceNumber.START_OF_TIME));
        }
    }
}

Example 8 with BookkeeperCommitLog

use of herddb.cluster.BookkeeperCommitLog in project herddb by diennea.

the class LedgerClosedTest method testLedgerClosedError.

@Test
public void testLedgerClosedError() throws Exception {
    ServerConfiguration serverconfig_1 = newServerConfigurationWithAutoPort(folder.newFolder().toPath());
    serverconfig_1.set(ServerConfiguration.PROPERTY_NODEID, "server1");
    serverconfig_1.set(ServerConfiguration.PROPERTY_MODE, ServerConfiguration.PROPERTY_MODE_CLUSTER);
    serverconfig_1.set(ServerConfiguration.PROPERTY_ZOOKEEPER_ADDRESS, testEnv.getAddress());
    serverconfig_1.set(ServerConfiguration.PROPERTY_ZOOKEEPER_PATH, testEnv.getPath());
    serverconfig_1.set(ServerConfiguration.PROPERTY_ZOOKEEPER_SESSIONTIMEOUT, testEnv.getTimeout());
    serverconfig_1.set(ServerConfiguration.PROPERTY_ENFORCE_LEADERSHIP, false);
    try (Server server = new Server(serverconfig_1)) {
        server.start();
        server.waitForStandaloneBoot();
        Table table = Table.builder().name("t1").column("c", ColumnTypes.INTEGER).primaryKey("c").build();
        server.getManager().executeStatement(new CreateTableStatement(table), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 1)), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 2)), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 3)), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        TableSpaceManager tableSpaceManager = server.getManager().getTableSpaceManager(TableSpace.DEFAULT);
        BookkeeperCommitLog log = (BookkeeperCommitLog) tableSpaceManager.getLog();
        long ledgerId = log.getLastSequenceNumber().ledgerId;
        assertTrue(ledgerId >= 0);
        // we do not want auto-recovery in ase of log failures
        server.getManager().setActivatorPauseStatus(true);
        assertEquals(ledgerId, log.getWriter().getOut().getId());
        // force close of the LedgerHandle
        // this may happen internally in BK in case of internal errors
        log.getWriter().getOut().close();
        // we should recover
        server.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 4)), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        assertNotEquals(ledgerId, log.getWriter().getOut().getId());
        ServerConfiguration serverconfig_2 = newServerConfigurationWithAutoPort(folder.newFolder().toPath());
        serverconfig_2.set(ServerConfiguration.PROPERTY_NODEID, "server2");
        serverconfig_2.set(ServerConfiguration.PROPERTY_MODE, ServerConfiguration.PROPERTY_MODE_CLUSTER);
        serverconfig_2.set(ServerConfiguration.PROPERTY_ZOOKEEPER_ADDRESS, testEnv.getAddress());
        serverconfig_2.set(ServerConfiguration.PROPERTY_ZOOKEEPER_PATH, testEnv.getPath());
        serverconfig_2.set(ServerConfiguration.PROPERTY_ZOOKEEPER_SESSIONTIMEOUT, testEnv.getTimeout());
        serverconfig_2.set(ServerConfiguration.PROPERTY_ENFORCE_LEADERSHIP, false);
        // set server2 as new leader
        server.getManager().executeStatement(new AlterTableSpaceStatement(TableSpace.DEFAULT, new HashSet<>(Arrays.asList("server1", "server2")), "server2", 1, 0), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        // stop server1
        server.close();
        // boot a new leader, it will recover from bookkeeper
        try (Server server2 = new Server(serverconfig_2)) {
            server2.start();
            // wait for the boot of the new leader
            server2.waitForTableSpaceBoot(TableSpace.DEFAULT, true);
            // the server must have all of the data
            try (DataScanner scan = scan(server2.getManager(), "SELECT * FROM t1", Collections.emptyList())) {
                List<DataAccessor> consume = scan.consume();
                assertEquals(4, consume.size());
            }
        }
    }
}

Example 9 with BookkeeperCommitLog

use of herddb.cluster.BookkeeperCommitLog in project herddb by diennea.

the class BootAsNewLeaderTest method testLeaderOnlineLogNoMoreAvailableDataAlreadyPresentBootAsNewLeader.

@Test
public void testLeaderOnlineLogNoMoreAvailableDataAlreadyPresentBootAsNewLeader() throws Exception {
    final AtomicInteger countErase = new AtomicInteger();
    SystemInstrumentation.addListener(new SystemInstrumentation.SingleInstrumentationPointListener("eraseTablespaceData") {

        @Override
        public void acceptSingle(Object... args) throws Exception {
            countErase.incrementAndGet();
        }
    });
    ServerConfiguration serverconfig_1 = newServerConfigurationWithAutoPort(folder.newFolder().toPath());
    serverconfig_1.set(ServerConfiguration.PROPERTY_NODEID, "server1");
    serverconfig_1.set(ServerConfiguration.PROPERTY_MODE, ServerConfiguration.PROPERTY_MODE_CLUSTER);
    serverconfig_1.set(ServerConfiguration.PROPERTY_ZOOKEEPER_ADDRESS, testEnv.getAddress());
    serverconfig_1.set(ServerConfiguration.PROPERTY_ZOOKEEPER_PATH, testEnv.getPath());
    serverconfig_1.set(ServerConfiguration.PROPERTY_ZOOKEEPER_SESSIONTIMEOUT, testEnv.getTimeout());
    serverconfig_1.set(ServerConfiguration.PROPERTY_ENFORCE_LEADERSHIP, false);
    serverconfig_1.set(ServerConfiguration.PROPERTY_BOOKKEEPER_LEDGERS_RETENTION_PERIOD, 1);
    serverconfig_1.set(ServerConfiguration.PROPERTY_CHECKPOINT_PERIOD, 0);
    // disabled
    serverconfig_1.set(ServerConfiguration.PROPERTY_BOOKKEEPER_MAX_IDLE_TIME, 0);
    ServerConfiguration serverconfig_2 = serverconfig_1.copy().set(ServerConfiguration.PROPERTY_NODEID, "server2").set(ServerConfiguration.PROPERTY_BASEDIR, folder.newFolder().toPath().toAbsolutePath());
    Table table = Table.builder().name("t1").column("c", ColumnTypes.INTEGER).column("s", ColumnTypes.INTEGER).primaryKey("c").build();
    Index index = Index.builder().onTable(table).type(Index.TYPE_BRIN).column("s", ColumnTypes.STRING).build();
    try (Server server_1 = new Server(serverconfig_1)) {
        server_1.start();
        server_1.waitForStandaloneBoot();
        server_1.getManager().executeStatement(new CreateTableStatement(table), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().executeStatement(new CreateIndexStatement(index), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 1, "s", "1")), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 2, "s", "2")), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 3, "s", "3")), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 4, "s", "4")), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().executeStatement(new AlterTableSpaceStatement(TableSpace.DEFAULT, new HashSet<>(Arrays.asList("server1", "server2")), "server1", 1, 0), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        TranslatedQuery translated = server_1.getManager().getPlanner().translate(TableSpace.DEFAULT, "SELECT * FROM " + TableSpace.DEFAULT + ".t1 WHERE s=1", Collections.emptyList(), true, true, false, -1);
        ScanStatement statement = translated.plan.mainStatement.unwrap(ScanStatement.class);
        assertTrue(statement.getPredicate().getIndexOperation() instanceof SecondaryIndexSeek);
        try (DataScanner scan = server_1.getManager().scan(statement, translated.context, TransactionContext.NO_TRANSACTION)) {
            assertEquals(1, scan.consume().size());
        }
    }
    String tableSpaceUUID;
    try (Server server_1 = new Server(serverconfig_1)) {
        server_1.start();
        server_1.waitForStandaloneBoot();
        {
            ZookeeperMetadataStorageManager man = (ZookeeperMetadataStorageManager) server_1.getMetadataStorageManager();
            tableSpaceUUID = man.describeTableSpace(TableSpace.DEFAULT).uuid;
            LedgersInfo ledgersList = ZookeeperMetadataStorageManager.readActualLedgersListFromZookeeper(man.getZooKeeper(), testEnv.getPath() + "/ledgers", tableSpaceUUID);
            assertEquals(2, ledgersList.getActiveLedgers().size());
        }
        server_1.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 5, "s", "5")), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        server_1.getManager().checkpoint();
    }
    try (Server server_1 = new Server(serverconfig_1)) {
        server_1.start();
        server_1.waitForStandaloneBoot();
        {
            ZookeeperMetadataStorageManager man = (ZookeeperMetadataStorageManager) server_1.getMetadataStorageManager();
            LedgersInfo ledgersList = ZookeeperMetadataStorageManager.readActualLedgersListFromZookeeper(man.getZooKeeper(), testEnv.getPath() + "/ledgers", tableSpaceUUID);
            assertEquals(2, ledgersList.getActiveLedgers().size());
        }
        server_1.getManager().executeUpdate(new InsertStatement(TableSpace.DEFAULT, "t1", RecordSerializer.makeRecord(table, "c", 6, "s", "6")), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        {
            ZookeeperMetadataStorageManager man = (ZookeeperMetadataStorageManager) server_1.getMetadataStorageManager();
            LedgersInfo ledgersList = ZookeeperMetadataStorageManager.readActualLedgersListFromZookeeper(man.getZooKeeper(), testEnv.getPath() + "/ledgers", tableSpaceUUID);
            assertEquals(2, ledgersList.getActiveLedgers().size());
        }
        server_1.getManager().checkpoint();
    }
    assertEquals(0, countErase.get());
    LogSequenceNumber server2checkpointPosition;
    try (Server server_1 = new Server(serverconfig_1)) {
        server_1.start();
        server_1.waitForStandaloneBoot();
        // start server_2, and flush data locally
        try (Server server_2 = new Server(serverconfig_2)) {
            server_2.start();
            assertTrue(server_2.getManager().waitForTablespace(TableSpace.DEFAULT, 60000, false));
            // wait for data to arrive on server_2
            for (int i = 0; i < 100; i++) {
                GetResult found = server_2.getManager().get(new GetStatement(TableSpace.DEFAULT, "t1", Bytes.from_int(1), null, false), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
                if (found.found()) {
                    break;
                }
                Thread.sleep(100);
            }
            // force a checkpoint, data is flushed to disk
            server_2.getManager().checkpoint();
            server2checkpointPosition = server_2.getManager().getTableSpaceManager(TableSpace.DEFAULT).getLog().getLastSequenceNumber();
            System.out.println("server2 checkpoint time: " + server2checkpointPosition);
        }
    }
    // start again server_1, in order to create a new ledger
    try (Server server_1 = new Server(serverconfig_1)) {
        server_1.start();
        server_1.waitForStandaloneBoot();
        BookkeeperCommitLog ll = (BookkeeperCommitLog) server_1.getManager().getTableSpaceManager(TableSpace.DEFAULT).getLog();
        // server_1 make much  progress
        ll.rollNewLedger();
        ll.rollNewLedger();
        ll.rollNewLedger();
        ll.rollNewLedger();
        ll.rollNewLedger();
        ll.rollNewLedger();
        ll.rollNewLedger();
        ll.rollNewLedger();
        // a checkpoint will delete old ledgers
        server_1.getManager().checkpoint();
        {
            ZookeeperMetadataStorageManager man = (ZookeeperMetadataStorageManager) server_1.getMetadataStorageManager();
            LedgersInfo ledgersList = ZookeeperMetadataStorageManager.readActualLedgersListFromZookeeper(man.getZooKeeper(), testEnv.getPath() + "/ledgers", tableSpaceUUID);
            System.out.println("ledgerList: " + ledgersList);
            assertEquals(1, ledgersList.getActiveLedgers().size());
            assertTrue(!ledgersList.getActiveLedgers().contains(ledgersList.getFirstLedger()));
            // we want to be sure that server_2 cannot recover from log
            assertTrue(!ledgersList.getActiveLedgers().contains(server2checkpointPosition.ledgerId));
        }
        assertEquals(1, countErase.get());
        // make server2 boot as leader, it will fence out server1
        // but server1 is not writing so it won't know from BK
        // it will see his new role from zookeeper (with a 'watch' notification)
        server_1.getManager().executeStatement(new AlterTableSpaceStatement(TableSpace.DEFAULT, new HashSet<>(Arrays.asList("server1", "server2")), "server2", 1, 0), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        // trying to become leader
        try (Server server_2 = new Server(serverconfig_2)) {
            server_2.start();
            assertFalse(server_2.getManager().waitForTablespace(TableSpace.DEFAULT, 10000, true));
        }
        // make server_1 leader again
        server_1.getManager().executeStatement(new AlterTableSpaceStatement(TableSpace.DEFAULT, new HashSet<>(Arrays.asList("server1", "server2")), "server1", 1, 0), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        // now server_2 must be able to boot, downloading data from server_1 and easing local state
        try (Server server_2 = new Server(serverconfig_2)) {
            server_2.start();
            assertFalse(server_2.getManager().waitForTablespace(TableSpace.DEFAULT, 60000, true));
        }
        // make server_2 leader again
        server_1.getManager().executeStatement(new AlterTableSpaceStatement(TableSpace.DEFAULT, new HashSet<>(Arrays.asList("server1", "server2")), "server2", 1, 0), StatementEvaluationContext.DEFAULT_EVALUATION_CONTEXT(), TransactionContext.NO_TRANSACTION);
        // stop server_1
        server_1.close();
        // it must have enough local data to boot
        try (Server server_2 = new Server(serverconfig_2)) {
            server_2.start();
            assertTrue(server_2.getManager().waitForTablespace(TableSpace.DEFAULT, 60000, true));
        }
    }
}

Example 10 with BookkeeperCommitLog

use of herddb.cluster.BookkeeperCommitLog in project herddb by diennea.

the class BookKeeperCommitLogTest method testMaxLedgerSize.

@Test
public void testMaxLedgerSize() throws Exception {
    final String tableSpaceUUID = UUID.randomUUID().toString();
    final String name = TableSpace.DEFAULT;
    final String nodeid = "nodeid";
    final int maxLedgerSize = 1024;
    final LogEntry entry = LogEntryFactory.beginTransaction(1);
    final int estimateSize = entry.serialize().length;
    final int numberOfLedgers = 10;
    final int numberOfEntries = 1 + numberOfLedgers * maxLedgerSize / estimateSize;
    ServerConfiguration serverConfiguration = newServerConfigurationWithAutoPort();
    try (ZookeeperMetadataStorageManager man = new ZookeeperMetadataStorageManager(testEnv.getAddress(), testEnv.getTimeout(), testEnv.getPath());
        BookkeeperCommitLogManager logManager = new BookkeeperCommitLogManager(man, serverConfiguration, NullStatsLogger.INSTANCE)) {
        logManager.setMaxLedgerSizeBytes(maxLedgerSize);
        man.start();
        logManager.start();
        try (BookkeeperCommitLog writer = logManager.createCommitLog(tableSpaceUUID, name, nodeid)) {
            // do not pollute the count with NOOP entries
            writer.setWriteLedgerHeader(false);
            writer.startWriting(1);
            for (int i = 0; i < numberOfEntries; i++) {
                writer.log(entry, false);
            }
            writer.log(entry, true).getLogSequenceNumber();
        }
        try (BookkeeperCommitLog reader = logManager.createCommitLog(tableSpaceUUID, name, nodeid)) {
            List<Map.Entry<LogSequenceNumber, LogEntry>> list = new ArrayList<>();
            Set<Long> ledgerIds = new HashSet<>();
            reader.recovery(LogSequenceNumber.START_OF_TIME, (a, b) -> {
                if (b.type != LogEntryType.NOOP) {
                    list.add(new AbstractMap.SimpleImmutableEntry<>(a, b));
                    ledgerIds.add(a.ledgerId);
                }
            }, false);
            assertEquals(numberOfEntries + 1, list.size());
            assertEquals(numberOfLedgers, ledgerIds.size());
        }
    }
}