Search in sources :

Example 1 with NamingThreadFactory

use of org.apache.accumulo.core.util.NamingThreadFactory in project accumulo by apache.

the class BloomFilterLayer method getLoadThreadPool.

private static synchronized ExecutorService getLoadThreadPool(int maxLoadThreads) {
    if (loadThreadPool != null) {
        return loadThreadPool;
    }
    if (maxLoadThreads > 0) {
        BlockingQueue<Runnable> q = new LinkedBlockingQueue<>();
        loadThreadPool = new ThreadPoolExecutor(0, maxLoadThreads, 60, TimeUnit.SECONDS, q, new NamingThreadFactory("bloom-loader"));
    }
    return loadThreadPool;
}
Also used : NamingThreadFactory(org.apache.accumulo.core.util.NamingThreadFactory) LoggingRunnable(org.apache.accumulo.fate.util.LoggingRunnable) ThreadPoolExecutor(java.util.concurrent.ThreadPoolExecutor) LinkedBlockingQueue(java.util.concurrent.LinkedBlockingQueue)

Example 2 with NamingThreadFactory

use of org.apache.accumulo.core.util.NamingThreadFactory in project accumulo by apache.

the class BulkImporter method assignMapFiles.

private Map<Path, List<KeyExtent>> assignMapFiles(String tableName, Map<Path, List<AssignmentInfo>> assignments, Map<KeyExtent, String> locations, int numThreads) {
    // group assignments by tablet
    Map<KeyExtent, List<PathSize>> assignmentsPerTablet = new TreeMap<>();
    for (Entry<Path, List<AssignmentInfo>> entry : assignments.entrySet()) {
        Path mapFile = entry.getKey();
        List<AssignmentInfo> tabletsToAssignMapFileTo = entry.getValue();
        for (AssignmentInfo ai : tabletsToAssignMapFileTo) {
            List<PathSize> mapFiles = assignmentsPerTablet.get(ai.ke);
            if (mapFiles == null) {
                mapFiles = new ArrayList<>();
                assignmentsPerTablet.put(ai.ke, mapFiles);
            }
            mapFiles.add(new PathSize(mapFile, ai.estSize));
        }
    }
    // group assignments by tabletserver
    Map<Path, List<KeyExtent>> assignmentFailures = Collections.synchronizedMap(new TreeMap<Path, List<KeyExtent>>());
    TreeMap<String, Map<KeyExtent, List<PathSize>>> assignmentsPerTabletServer = new TreeMap<>();
    for (Entry<KeyExtent, List<PathSize>> entry : assignmentsPerTablet.entrySet()) {
        KeyExtent ke = entry.getKey();
        String location = locations.get(ke);
        if (location == null) {
            for (PathSize pathSize : entry.getValue()) {
                synchronized (assignmentFailures) {
                    List<KeyExtent> failures = assignmentFailures.get(pathSize.path);
                    if (failures == null) {
                        failures = new ArrayList<>();
                        assignmentFailures.put(pathSize.path, failures);
                    }
                    failures.add(ke);
                }
            }
            log.warn("Could not assign {} map files to tablet {} because it had no location, will retry ...", entry.getValue().size(), ke);
            continue;
        }
        Map<KeyExtent, List<PathSize>> apt = assignmentsPerTabletServer.get(location);
        if (apt == null) {
            apt = new TreeMap<>();
            assignmentsPerTabletServer.put(location, apt);
        }
        apt.put(entry.getKey(), entry.getValue());
    }
    ExecutorService threadPool = Executors.newFixedThreadPool(numThreads, new NamingThreadFactory("submit"));
    for (Entry<String, Map<KeyExtent, List<PathSize>>> entry : assignmentsPerTabletServer.entrySet()) {
        String location = entry.getKey();
        threadPool.submit(new AssignmentTask(assignmentFailures, tableName, location, entry.getValue()));
    }
    threadPool.shutdown();
    while (!threadPool.isTerminated()) {
        try {
            threadPool.awaitTermination(60, TimeUnit.SECONDS);
        } catch (InterruptedException e) {
            log.error("Encountered InterruptedException while waiting for the thread pool to terminate.", e);
            throw new RuntimeException(e);
        }
    }
    return assignmentFailures;
}
Also used : NamingThreadFactory(org.apache.accumulo.core.util.NamingThreadFactory) TKeyExtent(org.apache.accumulo.core.data.thrift.TKeyExtent) KeyExtent(org.apache.accumulo.core.data.impl.KeyExtent) List(java.util.List) ArrayList(java.util.ArrayList) Path(org.apache.hadoop.fs.Path) TreeMap(java.util.TreeMap) ExecutorService(java.util.concurrent.ExecutorService) Map(java.util.Map) HashMap(java.util.HashMap) TreeMap(java.util.TreeMap)

Example 3 with NamingThreadFactory

use of org.apache.accumulo.core.util.NamingThreadFactory in project accumulo by apache.

the class BulkImporter method estimateSizes.

private Map<Path, List<AssignmentInfo>> estimateSizes(final AccumuloConfiguration acuConf, final Configuration conf, final VolumeManager vm, Map<Path, List<TabletLocation>> assignments, Collection<Path> paths, int numThreads) {
    long t1 = System.currentTimeMillis();
    final Map<Path, Long> mapFileSizes = new TreeMap<>();
    try {
        for (Path path : paths) {
            FileSystem fs = vm.getVolumeByPath(path).getFileSystem();
            mapFileSizes.put(path, fs.getContentSummary(path).getLength());
        }
    } catch (IOException e) {
        log.error("Failed to get map files in for {}: {}", paths, e.getMessage(), e);
        throw new RuntimeException(e);
    }
    final Map<Path, List<AssignmentInfo>> ais = Collections.synchronizedMap(new TreeMap<Path, List<AssignmentInfo>>());
    ExecutorService threadPool = Executors.newFixedThreadPool(numThreads, new NamingThreadFactory("estimateSizes"));
    for (final Entry<Path, List<TabletLocation>> entry : assignments.entrySet()) {
        if (entry.getValue().size() == 1) {
            TabletLocation tabletLocation = entry.getValue().get(0);
            // if the tablet completely contains the map file, there is no
            // need to estimate its
            // size
            ais.put(entry.getKey(), Collections.singletonList(new AssignmentInfo(tabletLocation.tablet_extent, mapFileSizes.get(entry.getKey()))));
            continue;
        }
        Runnable estimationTask = new Runnable() {

            @Override
            public void run() {
                Map<KeyExtent, Long> estimatedSizes = null;
                try {
                    estimatedSizes = FileUtil.estimateSizes(acuConf, entry.getKey(), mapFileSizes.get(entry.getKey()), extentsOf(entry.getValue()), conf, vm);
                } catch (IOException e) {
                    log.warn("Failed to estimate map file sizes {}", e.getMessage());
                }
                if (estimatedSizes == null) {
                    // estimation failed, do a simple estimation
                    estimatedSizes = new TreeMap<>();
                    long estSize = (long) (mapFileSizes.get(entry.getKey()) / (double) entry.getValue().size());
                    for (TabletLocation tl : entry.getValue()) estimatedSizes.put(tl.tablet_extent, estSize);
                }
                List<AssignmentInfo> assignmentInfoList = new ArrayList<>(estimatedSizes.size());
                for (Entry<KeyExtent, Long> entry2 : estimatedSizes.entrySet()) assignmentInfoList.add(new AssignmentInfo(entry2.getKey(), entry2.getValue()));
                ais.put(entry.getKey(), assignmentInfoList);
            }
        };
        threadPool.submit(new TraceRunnable(new LoggingRunnable(log, estimationTask)));
    }
    threadPool.shutdown();
    while (!threadPool.isTerminated()) {
        try {
            threadPool.awaitTermination(60, TimeUnit.SECONDS);
        } catch (InterruptedException e) {
            log.error("Encountered InterruptedException while waiting for the threadPool to terminate.", e);
            throw new RuntimeException(e);
        }
    }
    long t2 = System.currentTimeMillis();
    log.debug(String.format("Estimated map files sizes in %6.2f secs", (t2 - t1) / 1000.0));
    return ais;
}
Also used : NamingThreadFactory(org.apache.accumulo.core.util.NamingThreadFactory) ArrayList(java.util.ArrayList) TKeyExtent(org.apache.accumulo.core.data.thrift.TKeyExtent) KeyExtent(org.apache.accumulo.core.data.impl.KeyExtent) TraceRunnable(org.apache.htrace.wrappers.TraceRunnable) LoggingRunnable(org.apache.accumulo.fate.util.LoggingRunnable) TabletLocation(org.apache.accumulo.core.client.impl.TabletLocator.TabletLocation) FileSystem(org.apache.hadoop.fs.FileSystem) List(java.util.List) ArrayList(java.util.ArrayList) Path(org.apache.hadoop.fs.Path) IOException(java.io.IOException) TreeMap(java.util.TreeMap) TraceRunnable(org.apache.htrace.wrappers.TraceRunnable) LoggingRunnable(org.apache.accumulo.fate.util.LoggingRunnable) ExecutorService(java.util.concurrent.ExecutorService)

Example 4 with NamingThreadFactory

use of org.apache.accumulo.core.util.NamingThreadFactory in project accumulo by apache.

the class BulkImporter method importFiles.

public AssignmentStats importFiles(List<String> files, Path failureDir) throws IOException, AccumuloException, AccumuloSecurityException, ThriftTableOperationException {
    int numThreads = context.getConfiguration().getCount(Property.TSERV_BULK_PROCESS_THREADS);
    int numAssignThreads = context.getConfiguration().getCount(Property.TSERV_BULK_ASSIGNMENT_THREADS);
    timer = new StopWatch<>(Timers.class);
    timer.start(Timers.TOTAL);
    Configuration conf = CachedConfiguration.getInstance();
    VolumeManagerImpl.get(context.getConfiguration());
    final VolumeManager fs = VolumeManagerImpl.get(context.getConfiguration());
    Set<Path> paths = new HashSet<>();
    for (String file : files) {
        paths.add(new Path(file));
    }
    AssignmentStats assignmentStats = new AssignmentStats(paths.size());
    final Map<Path, List<KeyExtent>> completeFailures = Collections.synchronizedSortedMap(new TreeMap<Path, List<KeyExtent>>());
    ClientService.Client client = null;
    final TabletLocator locator = TabletLocator.getLocator(context, Table.ID.of(tableId));
    try {
        final Map<Path, List<TabletLocation>> assignments = Collections.synchronizedSortedMap(new TreeMap<Path, List<TabletLocation>>());
        timer.start(Timers.EXAMINE_MAP_FILES);
        ExecutorService threadPool = Executors.newFixedThreadPool(numThreads, new NamingThreadFactory("findOverlapping"));
        for (Path path : paths) {
            final Path mapFile = path;
            Runnable getAssignments = new Runnable() {

                @Override
                public void run() {
                    List<TabletLocation> tabletsToAssignMapFileTo = Collections.emptyList();
                    try {
                        tabletsToAssignMapFileTo = findOverlappingTablets(context, fs, locator, mapFile);
                    } catch (Exception ex) {
                        log.warn("Unable to find tablets that overlap file " + mapFile.toString(), ex);
                    }
                    log.debug("Map file {} found to overlap {} tablets", mapFile, tabletsToAssignMapFileTo.size());
                    if (tabletsToAssignMapFileTo.size() == 0) {
                        List<KeyExtent> empty = Collections.emptyList();
                        completeFailures.put(mapFile, empty);
                    } else
                        assignments.put(mapFile, tabletsToAssignMapFileTo);
                }
            };
            threadPool.submit(new TraceRunnable(new LoggingRunnable(log, getAssignments)));
        }
        threadPool.shutdown();
        while (!threadPool.isTerminated()) {
            try {
                threadPool.awaitTermination(60, TimeUnit.SECONDS);
            } catch (InterruptedException e) {
                throw new RuntimeException(e);
            }
        }
        timer.stop(Timers.EXAMINE_MAP_FILES);
        assignmentStats.attemptingAssignments(assignments);
        Map<Path, List<KeyExtent>> assignmentFailures = assignMapFiles(context, conf, fs, tableId, assignments, paths, numAssignThreads, numThreads);
        assignmentStats.assignmentsFailed(assignmentFailures);
        Map<Path, Integer> failureCount = new TreeMap<>();
        for (Entry<Path, List<KeyExtent>> entry : assignmentFailures.entrySet()) failureCount.put(entry.getKey(), 1);
        long sleepTime = 2 * 1000;
        while (assignmentFailures.size() > 0) {
            sleepTime = Math.min(sleepTime * 2, 60 * 1000);
            locator.invalidateCache();
            // assumption about assignment failures is that it caused by a split
            // happening or a missing location
            // 
            // for splits we need to find children key extents that cover the
            // same key range and are contiguous (no holes, no overlap)
            timer.start(Timers.SLEEP);
            sleepUninterruptibly(sleepTime, TimeUnit.MILLISECONDS);
            timer.stop(Timers.SLEEP);
            log.debug("Trying to assign {} map files that previously failed on some key extents", assignmentFailures.size());
            assignments.clear();
            // assign to
            for (Entry<Path, List<KeyExtent>> entry : assignmentFailures.entrySet()) {
                Iterator<KeyExtent> keListIter = entry.getValue().iterator();
                List<TabletLocation> tabletsToAssignMapFileTo = new ArrayList<>();
                while (keListIter.hasNext()) {
                    KeyExtent ke = keListIter.next();
                    timer.start(Timers.QUERY_METADATA);
                    try {
                        tabletsToAssignMapFileTo.addAll(findOverlappingTablets(context, fs, locator, entry.getKey(), ke));
                        keListIter.remove();
                    } catch (Exception ex) {
                        log.warn("Exception finding overlapping tablets, will retry tablet " + ke, ex);
                    }
                    timer.stop(Timers.QUERY_METADATA);
                }
                if (tabletsToAssignMapFileTo.size() > 0)
                    assignments.put(entry.getKey(), tabletsToAssignMapFileTo);
            }
            assignmentStats.attemptingAssignments(assignments);
            Map<Path, List<KeyExtent>> assignmentFailures2 = assignMapFiles(context, conf, fs, tableId, assignments, paths, numAssignThreads, numThreads);
            assignmentStats.assignmentsFailed(assignmentFailures2);
            // merge assignmentFailures2 into assignmentFailures
            for (Entry<Path, List<KeyExtent>> entry : assignmentFailures2.entrySet()) {
                assignmentFailures.get(entry.getKey()).addAll(entry.getValue());
                Integer fc = failureCount.get(entry.getKey());
                if (fc == null)
                    fc = 0;
                failureCount.put(entry.getKey(), fc + 1);
            }
            // remove map files that have no more key extents to assign
            Iterator<Entry<Path, List<KeyExtent>>> afIter = assignmentFailures.entrySet().iterator();
            while (afIter.hasNext()) {
                Entry<Path, List<KeyExtent>> entry = afIter.next();
                if (entry.getValue().size() == 0)
                    afIter.remove();
            }
            Set<Entry<Path, Integer>> failureIter = failureCount.entrySet();
            for (Entry<Path, Integer> entry : failureIter) {
                int retries = context.getConfiguration().getCount(Property.TSERV_BULK_RETRY);
                if (entry.getValue() > retries && assignmentFailures.get(entry.getKey()) != null) {
                    log.error("Map file {} failed more than {} times, giving up.", entry.getKey(), retries);
                    completeFailures.put(entry.getKey(), assignmentFailures.get(entry.getKey()));
                    assignmentFailures.remove(entry.getKey());
                }
            }
        }
        assignmentStats.assignmentsAbandoned(completeFailures);
        Set<Path> failedFailures = processFailures(completeFailures);
        assignmentStats.unrecoveredMapFiles(failedFailures);
        timer.stop(Timers.TOTAL);
        printReport(paths);
        return assignmentStats;
    } finally {
        if (client != null) {
            ServerClient.close(client);
        }
    }
}
Also used : VolumeManager(org.apache.accumulo.server.fs.VolumeManager) Configuration(org.apache.hadoop.conf.Configuration) AccumuloConfiguration(org.apache.accumulo.core.conf.AccumuloConfiguration) CachedConfiguration(org.apache.accumulo.core.util.CachedConfiguration) NamingThreadFactory(org.apache.accumulo.core.util.NamingThreadFactory) ArrayList(java.util.ArrayList) TKeyExtent(org.apache.accumulo.core.data.thrift.TKeyExtent) KeyExtent(org.apache.accumulo.core.data.impl.KeyExtent) TraceRunnable(org.apache.htrace.wrappers.TraceRunnable) LoggingRunnable(org.apache.accumulo.fate.util.LoggingRunnable) Entry(java.util.Map.Entry) TabletLocation(org.apache.accumulo.core.client.impl.TabletLocator.TabletLocation) List(java.util.List) ArrayList(java.util.ArrayList) HashSet(java.util.HashSet) Path(org.apache.hadoop.fs.Path) TabletClientService(org.apache.accumulo.core.tabletserver.thrift.TabletClientService) ClientService(org.apache.accumulo.core.client.impl.thrift.ClientService) TreeMap(java.util.TreeMap) ThriftSecurityException(org.apache.accumulo.core.client.impl.thrift.ThriftSecurityException) AccumuloSecurityException(org.apache.accumulo.core.client.AccumuloSecurityException) IOException(java.io.IOException) AccumuloException(org.apache.accumulo.core.client.AccumuloException) ThriftTableOperationException(org.apache.accumulo.core.client.impl.thrift.ThriftTableOperationException) TabletLocator(org.apache.accumulo.core.client.impl.TabletLocator) TraceRunnable(org.apache.htrace.wrappers.TraceRunnable) LoggingRunnable(org.apache.accumulo.fate.util.LoggingRunnable) ExecutorService(java.util.concurrent.ExecutorService)

Example 5 with NamingThreadFactory

use of org.apache.accumulo.core.util.NamingThreadFactory in project accumulo by apache.

the class TabletServerResourceManager method createIdlingEs.

private ExecutorService createIdlingEs(Property max, String name, long timeout, TimeUnit timeUnit) {
    LinkedBlockingQueue<Runnable> queue = new LinkedBlockingQueue<>();
    int maxThreads = conf.getSystemConfiguration().getCount(max);
    ThreadPoolExecutor tp = new ThreadPoolExecutor(maxThreads, maxThreads, timeout, timeUnit, queue, new NamingThreadFactory(name));
    tp.allowCoreThreadTimeOut(true);
    return addEs(max, name, tp);
}
Also used : NamingThreadFactory(org.apache.accumulo.core.util.NamingThreadFactory) LoggingRunnable(org.apache.accumulo.fate.util.LoggingRunnable) ThreadPoolExecutor(java.util.concurrent.ThreadPoolExecutor) LinkedBlockingQueue(java.util.concurrent.LinkedBlockingQueue)

Aggregations

NamingThreadFactory (org.apache.accumulo.core.util.NamingThreadFactory)8 ArrayList (java.util.ArrayList)4 ExecutorService (java.util.concurrent.ExecutorService)4 ThreadPoolExecutor (java.util.concurrent.ThreadPoolExecutor)4 LoggingRunnable (org.apache.accumulo.fate.util.LoggingRunnable)4 List (java.util.List)3 TreeMap (java.util.TreeMap)3 KeyExtent (org.apache.accumulo.core.data.impl.KeyExtent)3 TKeyExtent (org.apache.accumulo.core.data.thrift.TKeyExtent)3 Path (org.apache.hadoop.fs.Path)3 IOException (java.io.IOException)2 HashMap (java.util.HashMap)2 LinkedBlockingQueue (java.util.concurrent.LinkedBlockingQueue)2 AccumuloException (org.apache.accumulo.core.client.AccumuloException)2 AccumuloSecurityException (org.apache.accumulo.core.client.AccumuloSecurityException)2 TabletLocation (org.apache.accumulo.core.client.impl.TabletLocator.TabletLocation)2 TraceRunnable (org.apache.htrace.wrappers.TraceRunnable)2 PrintWriter (java.io.PrintWriter)1 StringWriter (java.io.StringWriter)1 HashSet (java.util.HashSet)1