Search in sources :

Example 1 with LoadQueuePeon

use of io.druid.server.coordinator.LoadQueuePeon in project druid by druid-io.

the class DruidCoordinatorCleanupUnneeded method run.

@Override
public DruidCoordinatorRuntimeParams run(DruidCoordinatorRuntimeParams params) {
    CoordinatorStats stats = new CoordinatorStats();
    Set<DataSegment> availableSegments = params.getAvailableSegments();
    DruidCluster cluster = params.getDruidCluster();
    // cleanup before it finished polling the metadata storage for available segments for the first time.
    if (!availableSegments.isEmpty()) {
        for (MinMaxPriorityQueue<ServerHolder> serverHolders : cluster.getSortedServersByTier()) {
            for (ServerHolder serverHolder : serverHolders) {
                ImmutableDruidServer server = serverHolder.getServer();
                for (ImmutableDruidDataSource dataSource : server.getDataSources()) {
                    for (DataSegment segment : dataSource.getSegments()) {
                        if (!availableSegments.contains(segment)) {
                            LoadQueuePeon queuePeon = params.getLoadManagementPeons().get(server.getName());
                            if (!queuePeon.getSegmentsToDrop().contains(segment)) {
                                queuePeon.dropSegment(segment, new LoadPeonCallback() {

                                    @Override
                                    public void execute() {
                                    }
                                });
                                stats.addToTieredStat("unneededCount", server.getTier(), 1);
                            }
                        }
                    }
                }
            }
        }
    } else {
        log.info("Found 0 availableSegments, skipping the cleanup of segments from historicals. This is done to prevent a race condition in which the coordinator would drop all segments if it started running cleanup before it finished polling the metadata storage for available segments for the first time.");
    }
    return params.buildFromExisting().withCoordinatorStats(stats).build();
}
Also used : CoordinatorStats(io.druid.server.coordinator.CoordinatorStats) LoadPeonCallback(io.druid.server.coordinator.LoadPeonCallback) ImmutableDruidDataSource(io.druid.client.ImmutableDruidDataSource) ServerHolder(io.druid.server.coordinator.ServerHolder) LoadQueuePeon(io.druid.server.coordinator.LoadQueuePeon) DruidCluster(io.druid.server.coordinator.DruidCluster) DataSegment(io.druid.timeline.DataSegment) ImmutableDruidServer(io.druid.client.ImmutableDruidServer)

Example 2 with LoadQueuePeon

use of io.druid.server.coordinator.LoadQueuePeon in project druid by druid-io.

the class DruidCoordinatorBalancer method moveSegment.

protected void moveSegment(final BalancerSegmentHolder segment, final ImmutableDruidServer toServer, final DruidCoordinatorRuntimeParams params) {
    final LoadQueuePeon toPeon = params.getLoadManagementPeons().get(toServer.getName());
    final ImmutableDruidServer fromServer = segment.getFromServer();
    final DataSegment segmentToMove = segment.getSegment();
    final String segmentName = segmentToMove.getIdentifier();
    if (!toPeon.getSegmentsToLoad().contains(segmentToMove) && (toServer.getSegment(segmentName) == null) && new ServerHolder(toServer, toPeon).getAvailableSize() > segmentToMove.getSize()) {
        log.info("Moving [%s] from [%s] to [%s]", segmentName, fromServer.getName(), toServer.getName());
        LoadPeonCallback callback = null;
        try {
            currentlyMovingSegments.get(toServer.getTier()).put(segmentName, segment);
            callback = new LoadPeonCallback() {

                @Override
                public void execute() {
                    Map<String, BalancerSegmentHolder> movingSegments = currentlyMovingSegments.get(toServer.getTier());
                    if (movingSegments != null) {
                        movingSegments.remove(segmentName);
                    }
                }
            };
            coordinator.moveSegment(fromServer, toServer, segmentToMove.getIdentifier(), callback);
        } catch (Exception e) {
            log.makeAlert(e, String.format("[%s] : Moving exception", segmentName)).emit();
            if (callback != null) {
                callback.execute();
            }
        }
    } else {
        currentlyMovingSegments.get(toServer.getTier()).remove(segmentName);
    }
}
Also used : LoadPeonCallback(io.druid.server.coordinator.LoadPeonCallback) ServerHolder(io.druid.server.coordinator.ServerHolder) LoadQueuePeon(io.druid.server.coordinator.LoadQueuePeon) DataSegment(io.druid.timeline.DataSegment) ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) Map(java.util.Map) ImmutableDruidServer(io.druid.client.ImmutableDruidServer)

Example 3 with LoadQueuePeon

use of io.druid.server.coordinator.LoadQueuePeon in project druid by druid-io.

the class DruidCoordinatorLogger method run.

@Override
public DruidCoordinatorRuntimeParams run(DruidCoordinatorRuntimeParams params) {
    DruidCluster cluster = params.getDruidCluster();
    CoordinatorStats stats = params.getCoordinatorStats();
    ServiceEmitter emitter = params.getEmitter();
    Map<String, AtomicLong> assigned = stats.getPerTierStats().get("assignedCount");
    if (assigned != null) {
        for (Map.Entry<String, AtomicLong> entry : assigned.entrySet()) {
            log.info("[%s] : Assigned %s segments among %,d servers", entry.getKey(), entry.getValue().get(), cluster.get(entry.getKey()).size());
        }
    }
    emitTieredStats(emitter, "segment/assigned/count", assigned);
    Map<String, AtomicLong> dropped = stats.getPerTierStats().get("droppedCount");
    if (dropped != null) {
        for (Map.Entry<String, AtomicLong> entry : dropped.entrySet()) {
            log.info("[%s] : Dropped %s segments among %,d servers", entry.getKey(), entry.getValue().get(), cluster.get(entry.getKey()).size());
        }
    }
    emitTieredStats(emitter, "segment/dropped/count", dropped);
    emitTieredStats(emitter, "segment/cost/raw", stats.getPerTierStats().get("initialCost"));
    emitTieredStats(emitter, "segment/cost/normalization", stats.getPerTierStats().get("normalization"));
    emitTieredStats(emitter, "segment/moved/count", stats.getPerTierStats().get("movedCount"));
    emitTieredStats(emitter, "segment/deleted/count", stats.getPerTierStats().get("deletedCount"));
    Map<String, AtomicLong> normalized = stats.getPerTierStats().get("normalizedInitialCostTimesOneThousand");
    if (normalized != null) {
        emitTieredStats(emitter, "segment/cost/normalized", Maps.transformEntries(normalized, new Maps.EntryTransformer<String, AtomicLong, Number>() {

            @Override
            public Number transformEntry(String key, AtomicLong value) {
                return value.doubleValue() / 1000d;
            }
        }));
    }
    Map<String, AtomicLong> unneeded = stats.getPerTierStats().get("unneededCount");
    if (unneeded != null) {
        for (Map.Entry<String, AtomicLong> entry : unneeded.entrySet()) {
            log.info("[%s] : Removed %s unneeded segments among %,d servers", entry.getKey(), entry.getValue().get(), cluster.get(entry.getKey()).size());
        }
    }
    emitTieredStats(emitter, "segment/unneeded/count", stats.getPerTierStats().get("unneededCount"));
    emitter.emit(new ServiceMetricEvent.Builder().build("segment/overShadowed/count", stats.getGlobalStats().get("overShadowedCount")));
    Map<String, AtomicLong> moved = stats.getPerTierStats().get("movedCount");
    if (moved != null) {
        for (Map.Entry<String, AtomicLong> entry : moved.entrySet()) {
            log.info("[%s] : Moved %,d segment(s)", entry.getKey(), entry.getValue().get());
        }
    }
    final Map<String, AtomicLong> unmoved = stats.getPerTierStats().get("unmovedCount");
    if (unmoved != null) {
        for (Map.Entry<String, AtomicLong> entry : unmoved.entrySet()) {
            log.info("[%s] : Let alone %,d segment(s)", entry.getKey(), entry.getValue().get());
        }
    }
    log.info("Load Queues:");
    for (MinMaxPriorityQueue<ServerHolder> serverHolders : cluster.getSortedServersByTier()) {
        for (ServerHolder serverHolder : serverHolders) {
            ImmutableDruidServer server = serverHolder.getServer();
            LoadQueuePeon queuePeon = serverHolder.getPeon();
            log.info("Server[%s, %s, %s] has %,d left to load, %,d left to drop, %,d bytes queued, %,d bytes served.", server.getName(), server.getType(), server.getTier(), queuePeon.getSegmentsToLoad().size(), queuePeon.getSegmentsToDrop().size(), queuePeon.getLoadQueueSize(), server.getCurrSize());
            if (log.isDebugEnabled()) {
                for (DataSegment segment : queuePeon.getSegmentsToLoad()) {
                    log.debug("Segment to load[%s]", segment);
                }
                for (DataSegment segment : queuePeon.getSegmentsToDrop()) {
                    log.debug("Segment to drop[%s]", segment);
                }
            }
        }
    }
    // Emit coordinator metrics
    final Set<Map.Entry<String, LoadQueuePeon>> peonEntries = params.getLoadManagementPeons().entrySet();
    for (Map.Entry<String, LoadQueuePeon> entry : peonEntries) {
        String serverName = entry.getKey();
        LoadQueuePeon queuePeon = entry.getValue();
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.SERVER, serverName).build("segment/loadQueue/size", queuePeon.getLoadQueueSize()));
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.SERVER, serverName).build("segment/loadQueue/failed", queuePeon.getAndResetFailedAssignCount()));
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.SERVER, serverName).build("segment/loadQueue/count", queuePeon.getSegmentsToLoad().size()));
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.SERVER, serverName).build("segment/dropQueue/count", queuePeon.getSegmentsToDrop().size()));
    }
    for (Map.Entry<String, AtomicLong> entry : coordinator.getSegmentAvailability().entrySet()) {
        String datasource = entry.getKey();
        Long count = entry.getValue().get();
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.DATASOURCE, datasource).build("segment/unavailable/count", count));
    }
    for (Map.Entry<String, CountingMap<String>> entry : coordinator.getReplicationStatus().entrySet()) {
        String tier = entry.getKey();
        CountingMap<String> datasourceAvailabilities = entry.getValue();
        for (Map.Entry<String, AtomicLong> datasourceAvailability : datasourceAvailabilities.entrySet()) {
            String datasource = datasourceAvailability.getKey();
            Long count = datasourceAvailability.getValue().get();
            emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.TIER, tier).setDimension(DruidMetrics.DATASOURCE, datasource).build("segment/underReplicated/count", count));
        }
    }
    // Emit segment metrics
    CountingMap<String> segmentSizes = new CountingMap<String>();
    CountingMap<String> segmentCounts = new CountingMap<String>();
    for (DruidDataSource dataSource : params.getDataSources()) {
        for (DataSegment segment : dataSource.getSegments()) {
            segmentSizes.add(dataSource.getName(), segment.getSize());
            segmentCounts.add(dataSource.getName(), 1L);
        }
    }
    for (Map.Entry<String, Long> entry : segmentSizes.snapshot().entrySet()) {
        String dataSource = entry.getKey();
        Long size = entry.getValue();
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.DATASOURCE, dataSource).build("segment/size", size));
    }
    for (Map.Entry<String, Long> entry : segmentCounts.snapshot().entrySet()) {
        String dataSource = entry.getKey();
        Long count = entry.getValue();
        emitter.emit(new ServiceMetricEvent.Builder().setDimension(DruidMetrics.DATASOURCE, dataSource).build("segment/count", count));
    }
    return params;
}
Also used : ServiceEmitter(com.metamx.emitter.service.ServiceEmitter) CoordinatorStats(io.druid.server.coordinator.CoordinatorStats) DataSegment(io.druid.timeline.DataSegment) LoadQueuePeon(io.druid.server.coordinator.LoadQueuePeon) DruidCluster(io.druid.server.coordinator.DruidCluster) DruidDataSource(io.druid.client.DruidDataSource) CountingMap(io.druid.collections.CountingMap) AtomicLong(java.util.concurrent.atomic.AtomicLong) ServerHolder(io.druid.server.coordinator.ServerHolder) AtomicLong(java.util.concurrent.atomic.AtomicLong) ServiceMetricEvent(com.metamx.emitter.service.ServiceMetricEvent) CountingMap(io.druid.collections.CountingMap) Map(java.util.Map) ImmutableDruidServer(io.druid.client.ImmutableDruidServer)

Aggregations

ImmutableDruidServer (io.druid.client.ImmutableDruidServer)3 LoadQueuePeon (io.druid.server.coordinator.LoadQueuePeon)3 ServerHolder (io.druid.server.coordinator.ServerHolder)3 DataSegment (io.druid.timeline.DataSegment)3 CoordinatorStats (io.druid.server.coordinator.CoordinatorStats)2 DruidCluster (io.druid.server.coordinator.DruidCluster)2 LoadPeonCallback (io.druid.server.coordinator.LoadPeonCallback)2 Map (java.util.Map)2 ServiceEmitter (com.metamx.emitter.service.ServiceEmitter)1 ServiceMetricEvent (com.metamx.emitter.service.ServiceMetricEvent)1 DruidDataSource (io.druid.client.DruidDataSource)1 ImmutableDruidDataSource (io.druid.client.ImmutableDruidDataSource)1 CountingMap (io.druid.collections.CountingMap)1 ConcurrentHashMap (java.util.concurrent.ConcurrentHashMap)1 AtomicLong (java.util.concurrent.atomic.AtomicLong)1