Examples with TServerConnection org.apache.accumulo.server.master.LiveTServerSet.TServerConnection used on opensource projects

Example 1 with TServerConnection

use of org.apache.accumulo.server.master.LiveTServerSet.TServerConnection in project accumulo by apache.

the class MasterClientServiceHandler method shutdownTabletServer.

@Override
public void shutdownTabletServer(TInfo info, TCredentials c, String tabletServer, boolean force) throws ThriftSecurityException {
    master.security.canPerformSystemActions(c);
    final TServerInstance doomed = master.tserverSet.find(tabletServer);
    if (!force) {
        final TServerConnection server = master.tserverSet.getConnection(doomed);
        if (server == null) {
            Master.log.warn("No server found for name {}", tabletServer);
            return;
        }
    }
    long tid = master.fate.startTransaction();
    log.debug("Seeding FATE op to shutdown " + tabletServer + " with tid " + tid);
    master.fate.seedTransaction(tid, new TraceRepo<>(new ShutdownTServer(doomed, force)), false);
    master.fate.waitForCompletion(tid);
    master.fate.delete(tid);
    log.debug("FATE op shutting down " + tabletServer + " finished");
}

Example 2 with TServerConnection

use of org.apache.accumulo.server.master.LiveTServerSet.TServerConnection in project accumulo by apache.

the class MasterClientServiceHandler method waitForFlush.

@Override
public void waitForFlush(TInfo tinfo, TCredentials c, String tableIdStr, ByteBuffer startRow, ByteBuffer endRow, long flushID, long maxLoops) throws ThriftSecurityException, ThriftTableOperationException {
    Table.ID tableId = Table.ID.of(tableIdStr);
    Namespace.ID namespaceId = getNamespaceIdFromTableId(TableOperation.FLUSH, tableId);
    master.security.canFlush(c, tableId, namespaceId);
    if (endRow != null && startRow != null && ByteBufferUtil.toText(startRow).compareTo(ByteBufferUtil.toText(endRow)) >= 0)
        throw new ThriftTableOperationException(tableId.canonicalID(), null, TableOperation.FLUSH, TableOperationExceptionType.BAD_RANGE, "start row must be less than end row");
    Set<TServerInstance> serversToFlush = new HashSet<>(master.tserverSet.getCurrentServers());
    for (long l = 0; l < maxLoops; l++) {
        for (TServerInstance instance : serversToFlush) {
            try {
                final TServerConnection server = master.tserverSet.getConnection(instance);
                if (server != null)
                    server.flush(master.masterLock, tableId, ByteBufferUtil.toBytes(startRow), ByteBufferUtil.toBytes(endRow));
            } catch (TException ex) {
                Master.log.error(ex.toString());
            }
        }
        if (l == maxLoops - 1)
            break;
        sleepUninterruptibly(50, TimeUnit.MILLISECONDS);
        serversToFlush.clear();
        try {
            Connector conn = master.getConnector();
            Scanner scanner;
            if (tableId.equals(MetadataTable.ID)) {
                scanner = new IsolatedScanner(conn.createScanner(RootTable.NAME, Authorizations.EMPTY));
                scanner.setRange(MetadataSchema.TabletsSection.getRange());
            } else {
                scanner = new IsolatedScanner(conn.createScanner(MetadataTable.NAME, Authorizations.EMPTY));
                Range range = new KeyExtent(tableId, null, ByteBufferUtil.toText(startRow)).toMetadataRange();
                scanner.setRange(range.clip(MetadataSchema.TabletsSection.getRange()));
            }
            TabletsSection.ServerColumnFamily.FLUSH_COLUMN.fetch(scanner);
            TabletsSection.ServerColumnFamily.DIRECTORY_COLUMN.fetch(scanner);
            scanner.fetchColumnFamily(TabletsSection.CurrentLocationColumnFamily.NAME);
            scanner.fetchColumnFamily(LogColumnFamily.NAME);
            RowIterator ri = new RowIterator(scanner);
            int tabletsToWaitFor = 0;
            int tabletCount = 0;
            Text ert = ByteBufferUtil.toText(endRow);
            while (ri.hasNext()) {
                Iterator<Entry<Key, Value>> row = ri.next();
                long tabletFlushID = -1;
                int logs = 0;
                boolean online = false;
                TServerInstance server = null;
                Entry<Key, Value> entry = null;
                while (row.hasNext()) {
                    entry = row.next();
                    Key key = entry.getKey();
                    if (TabletsSection.ServerColumnFamily.FLUSH_COLUMN.equals(key.getColumnFamily(), key.getColumnQualifier())) {
                        tabletFlushID = Long.parseLong(entry.getValue().toString());
                    }
                    if (LogColumnFamily.NAME.equals(key.getColumnFamily()))
                        logs++;
                    if (TabletsSection.CurrentLocationColumnFamily.NAME.equals(key.getColumnFamily())) {
                        online = true;
                        server = new TServerInstance(entry.getValue(), key.getColumnQualifier());
                    }
                }
                // when tablet is not online and has no logs, there is no reason to wait for it
                if ((online || logs > 0) && tabletFlushID < flushID) {
                    tabletsToWaitFor++;
                    if (server != null)
                        serversToFlush.add(server);
                }
                tabletCount++;
                Text tabletEndRow = new KeyExtent(entry.getKey().getRow(), (Text) null).getEndRow();
                if (tabletEndRow == null || (ert != null && tabletEndRow.compareTo(ert) >= 0))
                    break;
            }
            if (tabletsToWaitFor == 0)
                break;
            if (tabletCount == 0 && !Tables.exists(master.getInstance(), tableId))
                throw new ThriftTableOperationException(tableId.canonicalID(), null, TableOperation.FLUSH, TableOperationExceptionType.NOTFOUND, null);
        } catch (AccumuloException | TabletDeletedException e) {
            Master.log.debug("Failed to scan {} table to wait for flush {}", MetadataTable.NAME, tableId, e);
        } catch (AccumuloSecurityException e) {
            Master.log.warn("{}", e.getMessage(), e);
            throw new ThriftSecurityException();
        } catch (TableNotFoundException e) {
            Master.log.error("{}", e.getMessage(), e);
            throw new ThriftTableOperationException();
        }
    }
}

Example 3 with TServerConnection

use of org.apache.accumulo.server.master.LiveTServerSet.TServerConnection in project accumulo by apache.

the class LiveTServerSetTest method testSessionIds.

@Test
public void testSessionIds() {
    Map<String, TServerInfo> servers = new HashMap<>();
    TServerConnection mockConn = EasyMock.createMock(TServerConnection.class);
    TServerInfo server1 = new TServerInfo(new TServerInstance(HostAndPort.fromParts("localhost", 1234), "5555"), mockConn);
    servers.put("server1", server1);
    LiveTServerSet tservers = new LiveTServerSet(EasyMock.createMock(ClientContext.class), EasyMock.createMock(Listener.class));
    assertEquals(server1.instance, tservers.find(servers, "localhost:1234"));
    assertNull(tservers.find(servers, "localhost:4321"));
    assertEquals(server1.instance, tservers.find(servers, "localhost:1234[5555]"));
    assertNull(tservers.find(servers, "localhost:1234[55755]"));
}

Example 4 with TServerConnection

use of org.apache.accumulo.server.master.LiveTServerSet.TServerConnection in project accumulo by apache.

the class TabletGroupWatcher method run.

@Override
public void run() {
    Thread.currentThread().setName("Watching " + store.name());
    int[] oldCounts = new int[TabletState.values().length];
    EventCoordinator.Listener eventListener = this.master.nextEvent.getListener();
    WalStateManager wals = new WalStateManager(master.getInstance(), ZooReaderWriter.getInstance());
    while (this.master.stillMaster()) {
        // slow things down a little, otherwise we spam the logs when there are many wake-up events
        sleepUninterruptibly(100, TimeUnit.MILLISECONDS);
        masterState = master.getMasterState();
        int totalUnloaded = 0;
        int unloaded = 0;
        ClosableIterator<TabletLocationState> iter = null;
        try {
            Map<Table.ID, MergeStats> mergeStatsCache = new HashMap<>();
            Map<Table.ID, MergeStats> currentMerges = new HashMap<>();
            for (MergeInfo merge : master.merges()) {
                if (merge.getExtent() != null) {
                    currentMerges.put(merge.getExtent().getTableId(), new MergeStats(merge));
                }
            }
            // Get the current status for the current list of tservers
            SortedMap<TServerInstance, TabletServerStatus> currentTServers = new TreeMap<>();
            for (TServerInstance entry : this.master.tserverSet.getCurrentServers()) {
                currentTServers.put(entry, this.master.tserverStatus.get(entry));
            }
            if (currentTServers.size() == 0) {
                eventListener.waitForEvents(Master.TIME_TO_WAIT_BETWEEN_SCANS);
                synchronized (this) {
                    lastScanServers = ImmutableSortedSet.of();
                }
                continue;
            }
            // Don't move tablets to servers that are shutting down
            SortedMap<TServerInstance, TabletServerStatus> destinations = new TreeMap<>(currentTServers);
            destinations.keySet().removeAll(this.master.serversToShutdown);
            List<Assignment> assignments = new ArrayList<>();
            List<Assignment> assigned = new ArrayList<>();
            List<TabletLocationState> assignedToDeadServers = new ArrayList<>();
            List<TabletLocationState> suspendedToGoneServers = new ArrayList<>();
            Map<KeyExtent, TServerInstance> unassigned = new HashMap<>();
            Map<TServerInstance, List<Path>> logsForDeadServers = new TreeMap<>();
            MasterState masterState = master.getMasterState();
            int[] counts = new int[TabletState.values().length];
            stats.begin();
            // Walk through the tablets in our store, and work tablets
            // towards their goal
            iter = store.iterator();
            while (iter.hasNext()) {
                TabletLocationState tls = iter.next();
                if (tls == null) {
                    continue;
                }
                Master.log.debug("{} location State: {}", store.name(), tls);
                // ignore entries for tables that do not exist in zookeeper
                if (TableManager.getInstance().getTableState(tls.extent.getTableId()) == null)
                    continue;
                if (Master.log.isTraceEnabled())
                    Master.log.trace("{} walogs {}", tls, tls.walogs.size());
                // Don't overwhelm the tablet servers with work
                if (unassigned.size() + unloaded > Master.MAX_TSERVER_WORK_CHUNK * currentTServers.size()) {
                    flushChanges(destinations, assignments, assigned, assignedToDeadServers, logsForDeadServers, suspendedToGoneServers, unassigned);
                    assignments.clear();
                    assigned.clear();
                    assignedToDeadServers.clear();
                    suspendedToGoneServers.clear();
                    unassigned.clear();
                    unloaded = 0;
                    eventListener.waitForEvents(Master.TIME_TO_WAIT_BETWEEN_SCANS);
                }
                Table.ID tableId = tls.extent.getTableId();
                TableConfiguration tableConf = this.master.getConfigurationFactory().getTableConfiguration(tableId);
                MergeStats mergeStats = mergeStatsCache.get(tableId);
                if (mergeStats == null) {
                    mergeStats = currentMerges.get(tableId);
                    if (mergeStats == null) {
                        mergeStats = new MergeStats(new MergeInfo());
                    }
                    mergeStatsCache.put(tableId, mergeStats);
                }
                TabletGoalState goal = this.master.getGoalState(tls, mergeStats.getMergeInfo());
                TServerInstance server = tls.getServer();
                TabletState state = tls.getState(currentTServers.keySet());
                if (Master.log.isTraceEnabled()) {
                    Master.log.trace("Goal state {} current {} for {}", goal, state, tls.extent);
                }
                stats.update(tableId, state);
                mergeStats.update(tls.extent, state, tls.chopped, !tls.walogs.isEmpty());
                sendChopRequest(mergeStats.getMergeInfo(), state, tls);
                sendSplitRequest(mergeStats.getMergeInfo(), state, tls);
                // Always follow through with assignments
                if (state == TabletState.ASSIGNED) {
                    goal = TabletGoalState.HOSTED;
                }
                // if we are shutting down all the tabletservers, we have to do it in order
                if (goal == TabletGoalState.SUSPENDED && state == TabletState.HOSTED) {
                    if (this.master.serversToShutdown.equals(currentTServers.keySet())) {
                        if (dependentWatcher != null && dependentWatcher.assignedOrHosted() > 0) {
                            goal = TabletGoalState.HOSTED;
                        }
                    }
                }
                if (goal == TabletGoalState.HOSTED) {
                    if (state != TabletState.HOSTED && !tls.walogs.isEmpty()) {
                        if (this.master.recoveryManager.recoverLogs(tls.extent, tls.walogs))
                            continue;
                    }
                    switch(state) {
                        case HOSTED:
                            if (server.equals(this.master.migrations.get(tls.extent)))
                                this.master.migrations.remove(tls.extent);
                            break;
                        case ASSIGNED_TO_DEAD_SERVER:
                            assignedToDeadServers.add(tls);
                            if (server.equals(this.master.migrations.get(tls.extent)))
                                this.master.migrations.remove(tls.extent);
                            TServerInstance tserver = tls.futureOrCurrent();
                            if (!logsForDeadServers.containsKey(tserver)) {
                                logsForDeadServers.put(tserver, wals.getWalsInUse(tserver));
                            }
                            break;
                        case SUSPENDED:
                            if (master.getSteadyTime() - tls.suspend.suspensionTime < tableConf.getTimeInMillis(Property.TABLE_SUSPEND_DURATION)) {
                                // Tablet is suspended. See if its tablet server is back.
                                TServerInstance returnInstance = null;
                                Iterator<TServerInstance> find = destinations.tailMap(new TServerInstance(tls.suspend.server, " ")).keySet().iterator();
                                if (find.hasNext()) {
                                    TServerInstance found = find.next();
                                    if (found.getLocation().equals(tls.suspend.server)) {
                                        returnInstance = found;
                                    }
                                }
                                // Old tablet server is back. Return this tablet to its previous owner.
                                if (returnInstance != null) {
                                    assignments.add(new Assignment(tls.extent, returnInstance));
                                } else {
                                // leave suspended, don't ask for a new assignment.
                                }
                            } else {
                                // Treat as unassigned, ask for a new assignment.
                                unassigned.put(tls.extent, server);
                            }
                            break;
                        case UNASSIGNED:
                            // maybe it's a finishing migration
                            TServerInstance dest = this.master.migrations.get(tls.extent);
                            if (dest != null) {
                                // if destination is still good, assign it
                                if (destinations.keySet().contains(dest)) {
                                    assignments.add(new Assignment(tls.extent, dest));
                                } else {
                                    // get rid of this migration
                                    this.master.migrations.remove(tls.extent);
                                    unassigned.put(tls.extent, server);
                                }
                            } else {
                                unassigned.put(tls.extent, server);
                            }
                            break;
                        case ASSIGNED:
                            // Send another reminder
                            assigned.add(new Assignment(tls.extent, tls.future));
                            break;
                    }
                } else {
                    switch(state) {
                        case SUSPENDED:
                            // Request a move to UNASSIGNED, so as to allow balancing to continue.
                            suspendedToGoneServers.add(tls);
                            cancelOfflineTableMigrations(tls);
                            break;
                        case UNASSIGNED:
                            cancelOfflineTableMigrations(tls);
                            break;
                        case ASSIGNED_TO_DEAD_SERVER:
                            assignedToDeadServers.add(tls);
                            if (!logsForDeadServers.containsKey(tls.futureOrCurrent())) {
                                logsForDeadServers.put(tls.futureOrCurrent(), wals.getWalsInUse(tls.futureOrCurrent()));
                            }
                            break;
                        case HOSTED:
                            TServerConnection conn = this.master.tserverSet.getConnection(server);
                            if (conn != null) {
                                conn.unloadTablet(this.master.masterLock, tls.extent, goal.howUnload(), master.getSteadyTime());
                                unloaded++;
                                totalUnloaded++;
                            } else {
                                Master.log.warn("Could not connect to server {}", server);
                            }
                            break;
                        case ASSIGNED:
                            break;
                    }
                }
                counts[state.ordinal()]++;
            }
            flushChanges(destinations, assignments, assigned, assignedToDeadServers, logsForDeadServers, suspendedToGoneServers, unassigned);
            // provide stats after flushing changes to avoid race conditions w/ delete table
            stats.end(masterState);
            // Report changes
            for (TabletState state : TabletState.values()) {
                int i = state.ordinal();
                if (counts[i] > 0 && counts[i] != oldCounts[i]) {
                    this.master.nextEvent.event("[%s]: %d tablets are %s", store.name(), counts[i], state.name());
                }
            }
            Master.log.debug(String.format("[%s]: scan time %.2f seconds", store.name(), stats.getScanTime() / 1000.));
            oldCounts = counts;
            if (totalUnloaded > 0) {
                this.master.nextEvent.event("[%s]: %d tablets unloaded", store.name(), totalUnloaded);
            }
            updateMergeState(mergeStatsCache);
            synchronized (this) {
                lastScanServers = ImmutableSortedSet.copyOf(currentTServers.keySet());
            }
            if (this.master.tserverSet.getCurrentServers().equals(currentTServers.keySet())) {
                Master.log.debug(String.format("[%s] sleeping for %.2f seconds", store.name(), Master.TIME_TO_WAIT_BETWEEN_SCANS / 1000.));
                eventListener.waitForEvents(Master.TIME_TO_WAIT_BETWEEN_SCANS);
            } else {
                Master.log.info("Detected change in current tserver set, re-running state machine.");
            }
        } catch (Exception ex) {
            Master.log.error("Error processing table state for store " + store.name(), ex);
            if (ex.getCause() != null && ex.getCause() instanceof BadLocationStateException) {
                repairMetadata(((BadLocationStateException) ex.getCause()).getEncodedEndRow());
            } else {
                sleepUninterruptibly(Master.WAIT_BETWEEN_ERRORS, TimeUnit.MILLISECONDS);
            }
        } finally {
            if (iter != null) {
                try {
                    iter.close();
                } catch (IOException ex) {
                    Master.log.warn("Error closing TabletLocationState iterator: " + ex, ex);
                }
            }
        }
    }
}

Example 5 with TServerConnection

use of org.apache.accumulo.server.master.LiveTServerSet.TServerConnection in project accumulo by apache.

the class CopyFailed method isReady.

@Override
public long isReady(long tid, Master master) throws Exception {
    Set<TServerInstance> finished = new HashSet<>();
    Set<TServerInstance> running = master.onlineTabletServers();
    for (TServerInstance server : running) {
        try {
            TServerConnection client = master.getConnection(server);
            if (client != null && !client.isActive(tid))
                finished.add(server);
        } catch (TException ex) {
            log.info("Ignoring error trying to check on tid " + tid + " from server " + server + ": " + ex);
        }
    }
    if (finished.containsAll(running))
        return 0;
    return 500;
}