Search in sources :

Example 1 with TabletState

use of org.apache.accumulo.server.master.state.TabletState 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);
                }
            }
        }
    }
}
Also used : MergeInfo(org.apache.accumulo.server.master.state.MergeInfo) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) TabletGoalState(org.apache.accumulo.master.Master.TabletGoalState) KeyExtent(org.apache.accumulo.core.data.impl.KeyExtent) BadLocationStateException(org.apache.accumulo.server.master.state.TabletLocationState.BadLocationStateException) Assignment(org.apache.accumulo.server.master.state.Assignment) TabletLocationState(org.apache.accumulo.server.master.state.TabletLocationState) List(java.util.List) ArrayList(java.util.ArrayList) TabletServerStatus(org.apache.accumulo.core.master.thrift.TabletServerStatus) TableConfiguration(org.apache.accumulo.server.conf.TableConfiguration) MetadataTable(org.apache.accumulo.core.metadata.MetadataTable) RootTable(org.apache.accumulo.core.metadata.RootTable) Table(org.apache.accumulo.core.client.impl.Table) MasterState(org.apache.accumulo.core.master.thrift.MasterState) IOException(java.io.IOException) TreeMap(java.util.TreeMap) TServerInstance(org.apache.accumulo.server.master.state.TServerInstance) TableNotFoundException(org.apache.accumulo.core.client.TableNotFoundException) MutationsRejectedException(org.apache.accumulo.core.client.MutationsRejectedException) NotServingTabletException(org.apache.accumulo.core.tabletserver.thrift.NotServingTabletException) WalMarkerException(org.apache.accumulo.server.log.WalStateManager.WalMarkerException) AccumuloSecurityException(org.apache.accumulo.core.client.AccumuloSecurityException) DistributedStoreException(org.apache.accumulo.server.master.state.DistributedStoreException) TException(org.apache.thrift.TException) IOException(java.io.IOException) AccumuloException(org.apache.accumulo.core.client.AccumuloException) BadLocationStateException(org.apache.accumulo.server.master.state.TabletLocationState.BadLocationStateException) TServerConnection(org.apache.accumulo.server.master.LiveTServerSet.TServerConnection) TabletState(org.apache.accumulo.server.master.state.TabletState) WalStateManager(org.apache.accumulo.server.log.WalStateManager) MergeStats(org.apache.accumulo.master.state.MergeStats)

Example 2 with TabletState

use of org.apache.accumulo.server.master.state.TabletState in project accumulo by apache.

the class CleanUp method isReady.

@Override
public long isReady(long tid, Master master) throws Exception {
    if (!master.hasCycled(creationTime)) {
        return 50;
    }
    boolean done = true;
    Range tableRange = new KeyExtent(tableId, null, null).toMetadataRange();
    Scanner scanner = master.getConnector().createScanner(MetadataTable.NAME, Authorizations.EMPTY);
    MetaDataTableScanner.configureScanner(scanner, master);
    scanner.setRange(tableRange);
    for (Entry<Key, Value> entry : scanner) {
        TabletLocationState locationState = MetaDataTableScanner.createTabletLocationState(entry.getKey(), entry.getValue());
        TabletState state = locationState.getState(master.onlineTabletServers());
        if (state.equals(TabletState.ASSIGNED) || state.equals(TabletState.HOSTED)) {
            log.debug("Still waiting for table to be deleted: " + tableId + " locationState: " + locationState);
            done = false;
            break;
        }
    }
    if (!done)
        return 50;
    return 0;
}
Also used : BatchScanner(org.apache.accumulo.core.client.BatchScanner) MetaDataTableScanner(org.apache.accumulo.server.master.state.MetaDataTableScanner) Scanner(org.apache.accumulo.core.client.Scanner) TabletState(org.apache.accumulo.server.master.state.TabletState) Value(org.apache.accumulo.core.data.Value) TabletLocationState(org.apache.accumulo.server.master.state.TabletLocationState) Range(org.apache.accumulo.core.data.Range) KeyExtent(org.apache.accumulo.core.data.impl.KeyExtent) Key(org.apache.accumulo.core.data.Key)

Example 3 with TabletState

use of org.apache.accumulo.server.master.state.TabletState in project accumulo by apache.

the class FindOfflineTablets method checkTablets.

private static int checkTablets(Iterator<TabletLocationState> scanner, LiveTServerSet tservers) {
    int offline = 0;
    while (scanner.hasNext() && !System.out.checkError()) {
        TabletLocationState locationState = scanner.next();
        TabletState state = locationState.getState(tservers.getCurrentServers());
        if (state != null && state != TabletState.HOSTED && TableManager.getInstance().getTableState(locationState.extent.getTableId()) != TableState.OFFLINE) {
            System.out.println(locationState + " is " + state + "  #walogs:" + locationState.walogs.size());
            offline++;
        }
    }
    return offline;
}
Also used : TabletState(org.apache.accumulo.server.master.state.TabletState) TabletLocationState(org.apache.accumulo.server.master.state.TabletLocationState)

Aggregations

TabletLocationState (org.apache.accumulo.server.master.state.TabletLocationState)3 TabletState (org.apache.accumulo.server.master.state.TabletState)3 KeyExtent (org.apache.accumulo.core.data.impl.KeyExtent)2 IOException (java.io.IOException)1 ArrayList (java.util.ArrayList)1 HashMap (java.util.HashMap)1 List (java.util.List)1 TreeMap (java.util.TreeMap)1 AccumuloException (org.apache.accumulo.core.client.AccumuloException)1 AccumuloSecurityException (org.apache.accumulo.core.client.AccumuloSecurityException)1 BatchScanner (org.apache.accumulo.core.client.BatchScanner)1 MutationsRejectedException (org.apache.accumulo.core.client.MutationsRejectedException)1 Scanner (org.apache.accumulo.core.client.Scanner)1 TableNotFoundException (org.apache.accumulo.core.client.TableNotFoundException)1 Table (org.apache.accumulo.core.client.impl.Table)1 Key (org.apache.accumulo.core.data.Key)1 Range (org.apache.accumulo.core.data.Range)1 Value (org.apache.accumulo.core.data.Value)1 MasterState (org.apache.accumulo.core.master.thrift.MasterState)1 TabletServerStatus (org.apache.accumulo.core.master.thrift.TabletServerStatus)1