Search in sources :

Example 11 with TServerConnection

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

the class TabletGroupWatcher method run.

@Override
public void run() {
    int[] oldCounts = new int[TabletState.values().length];
    EventCoordinator.Listener eventListener = this.manager.nextEvent.getListener();
    WalStateManager wals = new WalStateManager(manager.getContext());
    while (manager.stillManager()) {
        // slow things down a little, otherwise we spam the logs when there are many wake-up events
        sleepUninterruptibly(100, TimeUnit.MILLISECONDS);
        int totalUnloaded = 0;
        int unloaded = 0;
        ClosableIterator<TabletLocationState> iter = null;
        try {
            Map<TableId, MergeStats> mergeStatsCache = new HashMap<>();
            Map<TableId, MergeStats> currentMerges = new HashMap<>();
            for (MergeInfo merge : manager.merges()) {
                if (merge.getExtent() != null) {
                    currentMerges.put(merge.getExtent().tableId(), new MergeStats(merge));
                }
            }
            // Get the current status for the current list of tservers
            SortedMap<TServerInstance, TabletServerStatus> currentTServers = new TreeMap<>();
            for (TServerInstance entry : manager.tserverSet.getCurrentServers()) {
                currentTServers.put(entry, manager.tserverStatus.get(entry));
            }
            if (currentTServers.isEmpty()) {
                eventListener.waitForEvents(Manager.TIME_TO_WAIT_BETWEEN_SCANS);
                synchronized (this) {
                    lastScanServers = Collections.emptySortedSet();
                }
                continue;
            }
            TabletLists tLists = new TabletLists(manager, currentTServers);
            ManagerState managerState = manager.getManagerState();
            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;
                }
                // ignore entries for tables that do not exist in zookeeper
                if (manager.getTableManager().getTableState(tls.extent.tableId()) == null)
                    continue;
                // Don't overwhelm the tablet servers with work
                if (tLists.unassigned.size() + unloaded > Manager.MAX_TSERVER_WORK_CHUNK * currentTServers.size()) {
                    flushChanges(tLists, wals);
                    tLists.reset();
                    unloaded = 0;
                    eventListener.waitForEvents(Manager.TIME_TO_WAIT_BETWEEN_SCANS);
                }
                TableId tableId = tls.extent.tableId();
                TableConfiguration tableConf = manager.getContext().getTableConfiguration(tableId);
                MergeStats mergeStats = mergeStatsCache.computeIfAbsent(tableId, k -> {
                    var mStats = currentMerges.get(k);
                    return mStats != null ? mStats : new MergeStats(new MergeInfo());
                });
                TabletGoalState goal = manager.getGoalState(tls, mergeStats.getMergeInfo());
                TServerInstance location = tls.getLocation();
                TabletState state = tls.getState(currentTServers.keySet());
                TabletLogger.missassigned(tls.extent, goal.toString(), state.toString(), tls.future, tls.current, tls.walogs.size());
                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) && manager.serversToShutdown.equals(currentTServers.keySet())) {
                    if (dependentWatcher != null && dependentWatcher.assignedOrHosted() > 0) {
                        goal = TabletGoalState.HOSTED;
                    }
                }
                if (goal == TabletGoalState.HOSTED) {
                    if ((state != TabletState.HOSTED && !tls.walogs.isEmpty()) && manager.recoveryManager.recoverLogs(tls.extent, tls.walogs))
                        continue;
                    switch(state) {
                        case HOSTED:
                            if (location.equals(manager.migrations.get(tls.extent)))
                                manager.migrations.remove(tls.extent);
                            break;
                        case ASSIGNED_TO_DEAD_SERVER:
                            hostDeadTablet(tLists, tls, location, wals);
                            break;
                        case SUSPENDED:
                            hostSuspendedTablet(tLists, tls, location, tableConf);
                            break;
                        case UNASSIGNED:
                            hostUnassignedTablet(tLists, tls.extent, location);
                            break;
                        case ASSIGNED:
                            // Send another reminder
                            tLists.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.
                            tLists.suspendedToGoneServers.add(tls);
                            cancelOfflineTableMigrations(tls.extent);
                            break;
                        case UNASSIGNED:
                            cancelOfflineTableMigrations(tls.extent);
                            break;
                        case ASSIGNED_TO_DEAD_SERVER:
                            unassignDeadTablet(tLists, tls, wals);
                            break;
                        case HOSTED:
                            TServerConnection client = manager.tserverSet.getConnection(location);
                            if (client != null) {
                                client.unloadTablet(manager.managerLock, tls.extent, goal.howUnload(), manager.getSteadyTime());
                                unloaded++;
                                totalUnloaded++;
                            } else {
                                Manager.log.warn("Could not connect to server {}", location);
                            }
                            break;
                        case ASSIGNED:
                            break;
                    }
                }
                counts[state.ordinal()]++;
            }
            flushChanges(tLists, wals);
            // provide stats after flushing changes to avoid race conditions w/ delete table
            stats.end(managerState);
            // Report changes
            for (TabletState state : TabletState.values()) {
                int i = state.ordinal();
                if (counts[i] > 0 && counts[i] != oldCounts[i]) {
                    manager.nextEvent.event("[%s]: %d tablets are %s", store.name(), counts[i], state.name());
                }
            }
            Manager.log.debug(String.format("[%s]: scan time %.2f seconds", store.name(), stats.getScanTime() / 1000.));
            oldCounts = counts;
            if (totalUnloaded > 0) {
                manager.nextEvent.event("[%s]: %d tablets unloaded", store.name(), totalUnloaded);
            }
            updateMergeState(mergeStatsCache);
            synchronized (this) {
                lastScanServers = ImmutableSortedSet.copyOf(currentTServers.keySet());
            }
            if (manager.tserverSet.getCurrentServers().equals(currentTServers.keySet())) {
                Manager.log.debug(String.format("[%s] sleeping for %.2f seconds", store.name(), Manager.TIME_TO_WAIT_BETWEEN_SCANS / 1000.));
                eventListener.waitForEvents(Manager.TIME_TO_WAIT_BETWEEN_SCANS);
            } else {
                Manager.log.info("Detected change in current tserver set, re-running state machine.");
            }
        } catch (Exception ex) {
            Manager.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(Manager.WAIT_BETWEEN_ERRORS, TimeUnit.MILLISECONDS);
            }
        } finally {
            if (iter != null) {
                try {
                    iter.close();
                } catch (IOException ex) {
                    Manager.log.warn("Error closing TabletLocationState iterator: " + ex, ex);
                }
            }
        }
    }
}
Also used : TableId(org.apache.accumulo.core.data.TableId) MergeInfo(org.apache.accumulo.server.manager.state.MergeInfo) HashMap(java.util.HashMap) TabletGoalState(org.apache.accumulo.manager.Manager.TabletGoalState) BadLocationStateException(org.apache.accumulo.core.metadata.TabletLocationState.BadLocationStateException) Assignment(org.apache.accumulo.server.manager.state.Assignment) ManagerState(org.apache.accumulo.core.manager.thrift.ManagerState) TabletLocationState(org.apache.accumulo.core.metadata.TabletLocationState) TabletServerStatus(org.apache.accumulo.core.master.thrift.TabletServerStatus) TableConfiguration(org.apache.accumulo.server.conf.TableConfiguration) IOException(java.io.IOException) TreeMap(java.util.TreeMap) TServerInstance(org.apache.accumulo.core.metadata.TServerInstance) TableNotFoundException(org.apache.accumulo.core.client.TableNotFoundException) DistributedStoreException(org.apache.accumulo.server.manager.state.DistributedStoreException) MutationsRejectedException(org.apache.accumulo.core.client.MutationsRejectedException) NotServingTabletException(org.apache.accumulo.core.tabletserver.thrift.NotServingTabletException) WalMarkerException(org.apache.accumulo.server.log.WalStateManager.WalMarkerException) BadLocationStateException(org.apache.accumulo.core.metadata.TabletLocationState.BadLocationStateException) TException(org.apache.thrift.TException) IOException(java.io.IOException) AccumuloException(org.apache.accumulo.core.client.AccumuloException) TServerConnection(org.apache.accumulo.server.manager.LiveTServerSet.TServerConnection) TabletState(org.apache.accumulo.core.metadata.TabletState) WalStateManager(org.apache.accumulo.server.log.WalStateManager) MergeStats(org.apache.accumulo.manager.state.MergeStats)

Example 12 with TServerConnection

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

the class TabletGroupWatcher method sendChopRequest.

private void sendChopRequest(MergeInfo info, TabletState state, TabletLocationState tls) {
    // Don't bother if we're in the wrong state
    if (!info.getState().equals(MergeState.WAITING_FOR_CHOPPED))
        return;
    // Tablet must be online
    if (!state.equals(TabletState.HOSTED))
        return;
    // Tablet isn't already chopped
    if (tls.chopped)
        return;
    // Tablet ranges intersect
    if (info.needsToBeChopped(tls.extent)) {
        TServerConnection conn;
        try {
            conn = manager.tserverSet.getConnection(tls.current);
            if (conn != null) {
                Manager.log.info("Asking {} to chop {}", tls.current, tls.extent);
                conn.chop(manager.managerLock, tls.extent);
            } else {
                Manager.log.warn("Could not connect to server {}", tls.current);
            }
        } catch (TException e) {
            Manager.log.warn("Communications error asking tablet server to chop a tablet");
        }
    }
}
Also used : TServerConnection(org.apache.accumulo.server.manager.LiveTServerSet.TServerConnection) TException(org.apache.thrift.TException)

Example 13 with TServerConnection

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

the class TabletGroupWatcher method flushChanges.

private void flushChanges(TabletLists tLists, WalStateManager wals) throws DistributedStoreException, TException, WalMarkerException {
    var unassigned = Collections.unmodifiableMap(tLists.unassigned);
    handleDeadTablets(tLists, wals);
    getAssignmentsFromBalancer(tLists, unassigned);
    if (!tLists.assignments.isEmpty()) {
        Manager.log.info(String.format("Assigning %d tablets", tLists.assignments.size()));
        store.setFutureLocations(tLists.assignments);
    }
    tLists.assignments.addAll(tLists.assigned);
    for (Assignment a : tLists.assignments) {
        TServerConnection client = manager.tserverSet.getConnection(a.server);
        if (client != null) {
            client.assignTablet(manager.managerLock, a.tablet);
        } else {
            Manager.log.warn("Could not connect to server {}", a.server);
        }
        manager.assignedTablet(a.tablet);
    }
}
Also used : Assignment(org.apache.accumulo.server.manager.state.Assignment) TServerConnection(org.apache.accumulo.server.manager.LiveTServerSet.TServerConnection)

Aggregations

TServerConnection (org.apache.accumulo.server.manager.LiveTServerSet.TServerConnection)13 TServerInstance (org.apache.accumulo.core.metadata.TServerInstance)9 TException (org.apache.thrift.TException)7 TabletServerStatus (org.apache.accumulo.core.master.thrift.TabletServerStatus)4 IOException (java.io.IOException)3 HashMap (java.util.HashMap)3 HashSet (java.util.HashSet)3 TableNotFoundException (org.apache.accumulo.core.client.TableNotFoundException)3 TableId (org.apache.accumulo.core.data.TableId)3 ServerContext (org.apache.accumulo.server.ServerContext)3 TreeMap (java.util.TreeMap)2 ThriftSecurityException (org.apache.accumulo.core.clientImpl.thrift.ThriftSecurityException)2 ThriftTableOperationException (org.apache.accumulo.core.clientImpl.thrift.ThriftTableOperationException)2 KeyExtent (org.apache.accumulo.core.dataImpl.KeyExtent)2 TableInfo (org.apache.accumulo.core.master.thrift.TableInfo)2 TabletMetadata (org.apache.accumulo.core.metadata.schema.TabletMetadata)2 TabletsMetadata (org.apache.accumulo.core.metadata.schema.TabletsMetadata)2 ZooReaderWriter (org.apache.accumulo.fate.zookeeper.ZooReaderWriter)2 ShutdownTServer (org.apache.accumulo.manager.tserverOps.ShutdownTServer)2 Assignment (org.apache.accumulo.server.manager.state.Assignment)2