Examples with KeyExtent org.apache.accumulo.core.dataImpl.KeyExtent used on opensource projects

Example 21 with KeyExtent

use of org.apache.accumulo.core.dataImpl.KeyExtent in project accumulo by apache.

the class BulkImport method computeFileToTabletMappings.

public SortedMap<KeyExtent, Bulk.Files> computeFileToTabletMappings(FileSystem fs, TableId tableId, Path dirPath, Executor executor, ClientContext context, int maxTablets) throws IOException {
    KeyExtentCache extentCache = new ConcurrentKeyExtentCache(tableId, context);
    List<FileStatus> files = filterInvalid(fs.listStatus(dirPath, p -> !p.getName().equals(Constants.BULK_LOAD_MAPPING)));
    // we know all of the file lens, so construct a cache and populate it in order to avoid later
    // trips to the namenode
    Cache<String, Long> fileLensCache = getPopulatedFileLenCache(dirPath, files);
    List<CompletableFuture<Map<KeyExtent, Bulk.FileInfo>>> futures = new ArrayList<>();
    CryptoService cs = CryptoServiceFactory.newDefaultInstance();
    for (FileStatus fileStatus : files) {
        Path filePath = fileStatus.getPath();
        CompletableFuture<Map<KeyExtent, Bulk.FileInfo>> future = CompletableFuture.supplyAsync(() -> {
            try {
                long t1 = System.currentTimeMillis();
                List<KeyExtent> extents = findOverlappingTablets(context, extentCache, filePath, fs, fileLensCache, cs);
                // make sure file isn't going to too many tablets
                checkTabletCount(maxTablets, extents.size(), filePath.toString());
                Map<KeyExtent, Long> estSizes = estimateSizes(context.getConfiguration(), filePath, fileStatus.getLen(), extents, fs, fileLensCache, cs);
                Map<KeyExtent, Bulk.FileInfo> pathLocations = new HashMap<>();
                for (KeyExtent ke : extents) {
                    pathLocations.put(ke, new Bulk.FileInfo(filePath, estSizes.getOrDefault(ke, 0L)));
                }
                long t2 = System.currentTimeMillis();
                log.debug("Mapped {} to {} tablets in {}ms", filePath, pathLocations.size(), t2 - t1);
                return pathLocations;
            } catch (Exception e) {
                throw new CompletionException(e);
            }
        }, executor);
        futures.add(future);
    }
    SortedMap<KeyExtent, Bulk.Files> mappings = new TreeMap<>();
    for (CompletableFuture<Map<KeyExtent, Bulk.FileInfo>> future : futures) {
        try {
            Map<KeyExtent, Bulk.FileInfo> pathMapping = future.get();
            pathMapping.forEach((ext, fi) -> mappings.computeIfAbsent(ext, k -> new Files()).add(fi));
        } catch (InterruptedException e) {
            Thread.currentThread().interrupt();
            throw new RuntimeException(e);
        } catch (ExecutionException e) {
            throw new RuntimeException(e);
        }
    }
    return mergeOverlapping(mappings);
}

Example 22 with KeyExtent

use of org.apache.accumulo.core.dataImpl.KeyExtent in project accumulo by apache.

the class BulkImport method findOverlappingTablets.

public static List<KeyExtent> findOverlappingTablets(KeyExtentCache extentCache, FileSKVIterator reader) throws IOException {
    List<KeyExtent> result = new ArrayList<>();
    Collection<ByteSequence> columnFamilies = Collections.emptyList();
    Text row = new Text();
    while (true) {
        reader.seek(new Range(row, null), columnFamilies, false);
        if (!reader.hasTop()) {
            break;
        }
        row = reader.getTopKey().getRow();
        KeyExtent extent = extentCache.lookup(row);
        result.add(extent);
        row = extent.endRow();
        if (row != null) {
            row = nextRow(row);
        } else
            break;
    }
    return result;
}

Example 23 with KeyExtent

use of org.apache.accumulo.core.dataImpl.KeyExtent in project accumulo by apache.

the class TabletServerBatchReaderIterator method doLookup.

static void doLookup(ClientContext context, String server, Map<KeyExtent, List<Range>> requested, Map<KeyExtent, List<Range>> failures, Map<KeyExtent, List<Range>> unscanned, ResultReceiver receiver, List<Column> columns, ScannerOptions options, Authorizations authorizations, TimeoutTracker timeoutTracker) throws IOException, AccumuloSecurityException, AccumuloServerException {
    if (requested.isEmpty()) {
        return;
    }
    // copy requested to unscanned map. we will remove ranges as they are scanned in trackScanning()
    for (Entry<KeyExtent, List<Range>> entry : requested.entrySet()) {
        ArrayList<Range> ranges = new ArrayList<>();
        for (Range range : entry.getValue()) {
            ranges.add(new Range(range));
        }
        unscanned.put(KeyExtent.copyOf(entry.getKey()), ranges);
    }
    timeoutTracker.startingScan();
    try {
        final HostAndPort parsedServer = HostAndPort.fromString(server);
        final TabletClientService.Client client;
        if (timeoutTracker.getTimeOut() < context.getClientTimeoutInMillis())
            client = ThriftUtil.getTServerClient(parsedServer, context, timeoutTracker.getTimeOut());
        else
            client = ThriftUtil.getTServerClient(parsedServer, context);
        try {
            OpTimer timer = null;
            if (log.isTraceEnabled()) {
                log.trace("tid={} Starting multi scan, tserver={}  #tablets={}  #ranges={} ssil={} ssio={}", Thread.currentThread().getId(), server, requested.size(), sumSizes(requested.values()), options.serverSideIteratorList, options.serverSideIteratorOptions);
                timer = new OpTimer().start();
            }
            TabletType ttype = TabletType.type(requested.keySet());
            boolean waitForWrites = !ThriftScanner.serversWaitedForWrites.get(ttype).contains(server);
            // @formatter:off
            Map<TKeyExtent, List<TRange>> thriftTabletRanges = requested.entrySet().stream().collect(Collectors.toMap((entry) -> entry.getKey().toThrift(), (entry) -> entry.getValue().stream().map(Range::toThrift).collect(Collectors.toList())));
            // @formatter:on
            Map<String, String> execHints = options.executionHints.isEmpty() ? null : options.executionHints;
            InitialMultiScan imsr = client.startMultiScan(TraceUtil.traceInfo(), context.rpcCreds(), thriftTabletRanges, columns.stream().map(Column::toThrift).collect(Collectors.toList()), options.serverSideIteratorList, options.serverSideIteratorOptions, ByteBufferUtil.toByteBuffers(authorizations.getAuthorizations()), waitForWrites, SamplerConfigurationImpl.toThrift(options.getSamplerConfiguration()), options.batchTimeOut, options.classLoaderContext, execHints);
            if (waitForWrites)
                ThriftScanner.serversWaitedForWrites.get(ttype).add(server.toString());
            MultiScanResult scanResult = imsr.result;
            if (timer != null) {
                timer.stop();
                log.trace("tid={} Got 1st multi scan results, #results={} {} in {}", Thread.currentThread().getId(), scanResult.results.size(), (scanResult.more ? "scanID=" + imsr.scanID : ""), String.format("%.3f secs", timer.scale(SECONDS)));
            }
            ArrayList<Entry<Key, Value>> entries = new ArrayList<>(scanResult.results.size());
            for (TKeyValue kv : scanResult.results) {
                entries.add(new SimpleImmutableEntry<>(new Key(kv.key), new Value(kv.value)));
            }
            if (!entries.isEmpty())
                receiver.receive(entries);
            if (!entries.isEmpty() || !scanResult.fullScans.isEmpty())
                timeoutTracker.madeProgress();
            trackScanning(failures, unscanned, scanResult);
            AtomicLong nextOpid = new AtomicLong();
            while (scanResult.more) {
                timeoutTracker.check();
                if (timer != null) {
                    log.trace("tid={} oid={} Continuing multi scan, scanid={}", Thread.currentThread().getId(), nextOpid.get(), imsr.scanID);
                    timer.reset().start();
                }
                scanResult = client.continueMultiScan(TraceUtil.traceInfo(), imsr.scanID);
                if (timer != null) {
                    timer.stop();
                    log.trace("tid={} oid={} Got more multi scan results, #results={} {} in {}", Thread.currentThread().getId(), nextOpid.getAndIncrement(), scanResult.results.size(), (scanResult.more ? " scanID=" + imsr.scanID : ""), String.format("%.3f secs", timer.scale(SECONDS)));
                }
                entries = new ArrayList<>(scanResult.results.size());
                for (TKeyValue kv : scanResult.results) {
                    entries.add(new SimpleImmutableEntry<>(new Key(kv.key), new Value(kv.value)));
                }
                if (!entries.isEmpty())
                    receiver.receive(entries);
                if (!entries.isEmpty() || !scanResult.fullScans.isEmpty())
                    timeoutTracker.madeProgress();
                trackScanning(failures, unscanned, scanResult);
            }
            client.closeMultiScan(TraceUtil.traceInfo(), imsr.scanID);
        } finally {
            ThriftUtil.returnClient(client, context);
        }
    } catch (TTransportException e) {
        log.debug("Server : {} msg : {}", server, e.getMessage());
        timeoutTracker.errorOccured();
        throw new IOException(e);
    } catch (ThriftSecurityException e) {
        log.debug("Server : {} msg : {}", server, e.getMessage(), e);
        throw new AccumuloSecurityException(e.user, e.code, e);
    } catch (TApplicationException e) {
        log.debug("Server : {} msg : {}", server, e.getMessage(), e);
        throw new AccumuloServerException(server, e);
    } catch (NoSuchScanIDException e) {
        log.debug("Server : {} msg : {}", server, e.getMessage(), e);
        throw new IOException(e);
    } catch (TSampleNotPresentException e) {
        log.debug("Server : " + server + " msg : " + e.getMessage(), e);
        String tableInfo = "?";
        if (e.getExtent() != null) {
            TableId tableId = KeyExtent.fromThrift(e.getExtent()).tableId();
            tableInfo = context.getPrintableTableInfoFromId(tableId);
        }
        String message = "Table " + tableInfo + " does not have sampling configured or built";
        throw new SampleNotPresentException(message, e);
    } catch (TException e) {
        log.debug("Server : {} msg : {}", server, e.getMessage(), e);
        timeoutTracker.errorOccured();
        throw new IOException(e);
    }
}

Example 24 with KeyExtent

use of org.apache.accumulo.core.dataImpl.KeyExtent in project accumulo by apache.

the class TabletServerBatchReaderIterator method doLookups.

private void doLookups(Map<String, Map<KeyExtent, List<Range>>> binnedRanges, final ResultReceiver receiver, List<Column> columns) {
    if (timedoutServers.containsAll(binnedRanges.keySet())) {
        // all servers have timed out
        throw new TimedOutException(timedoutServers);
    }
    // when there are lots of threads and a few tablet servers
    // it is good to break request to tablet servers up, the
    // following code determines if this is the case
    int maxTabletsPerRequest = Integer.MAX_VALUE;
    if (numThreads / binnedRanges.size() > 1) {
        int totalNumberOfTablets = 0;
        for (Entry<String, Map<KeyExtent, List<Range>>> entry : binnedRanges.entrySet()) {
            totalNumberOfTablets += entry.getValue().size();
        }
        maxTabletsPerRequest = totalNumberOfTablets / numThreads;
        if (maxTabletsPerRequest == 0) {
            maxTabletsPerRequest = 1;
        }
    }
    Map<KeyExtent, List<Range>> failures = new HashMap<>();
    if (!timedoutServers.isEmpty()) {
        // go ahead and fail any timed out servers
        for (Iterator<Entry<String, Map<KeyExtent, List<Range>>>> iterator = binnedRanges.entrySet().iterator(); iterator.hasNext(); ) {
            Entry<String, Map<KeyExtent, List<Range>>> entry = iterator.next();
            if (timedoutServers.contains(entry.getKey())) {
                failures.putAll(entry.getValue());
                iterator.remove();
            }
        }
    }
    // randomize tabletserver order... this will help when there are multiple
    // batch readers and writers running against accumulo
    List<String> locations = new ArrayList<>(binnedRanges.keySet());
    Collections.shuffle(locations);
    List<QueryTask> queryTasks = new ArrayList<>();
    for (final String tsLocation : locations) {
        final Map<KeyExtent, List<Range>> tabletsRanges = binnedRanges.get(tsLocation);
        if (maxTabletsPerRequest == Integer.MAX_VALUE || tabletsRanges.size() == 1) {
            QueryTask queryTask = new QueryTask(tsLocation, tabletsRanges, failures, receiver, columns);
            queryTasks.add(queryTask);
        } else {
            HashMap<KeyExtent, List<Range>> tabletSubset = new HashMap<>();
            for (Entry<KeyExtent, List<Range>> entry : tabletsRanges.entrySet()) {
                tabletSubset.put(entry.getKey(), entry.getValue());
                if (tabletSubset.size() >= maxTabletsPerRequest) {
                    QueryTask queryTask = new QueryTask(tsLocation, tabletSubset, failures, receiver, columns);
                    queryTasks.add(queryTask);
                    tabletSubset = new HashMap<>();
                }
            }
            if (!tabletSubset.isEmpty()) {
                QueryTask queryTask = new QueryTask(tsLocation, tabletSubset, failures, receiver, columns);
                queryTasks.add(queryTask);
            }
        }
    }
    final Semaphore semaphore = new Semaphore(queryTasks.size());
    semaphore.acquireUninterruptibly(queryTasks.size());
    for (QueryTask queryTask : queryTasks) {
        queryTask.setSemaphore(semaphore, queryTasks.size());
        queryThreadPool.execute(queryTask);
    }
}

Example 25 with KeyExtent

use of org.apache.accumulo.core.dataImpl.KeyExtent in project accumulo by apache.

the class ConcurrentKeyExtentCache method lookup.

@Override
public KeyExtent lookup(Text row) {
    while (true) {
        KeyExtent ke = getFromCache(row);
        if (ke != null)
            return ke;
        // If a metadata lookup is currently in progress, then multiple threads can queue up their
        // rows. The next lookup will process all queued. Processing multiple at once can be more
        // efficient.
        rowsToLookup.add(row);
        synchronized (this) {
            // This check is done to avoid processing rowsToLookup when the current thread's row is in
            // the cache.
            ke = getFromCache(row);
            if (ke != null) {
                rowsToLookup.remove(row);
                return ke;
            }
            lookupRows.clear();
            synchronized (rowsToLookup) {
                // Gather all rows that were queued for lookup before this point in time.
                rowsToLookup.forEach(lookupRows::add);
                rowsToLookup.clear();
            }
            // Lookup rows in the metadata table in sorted order. This could possibly lead to less
            // metadata lookups.
            lookupRows.sort(Text::compareTo);
            for (Text lookupRow : lookupRows) {
                if (getFromCache(lookupRow) == null) {
                    Iterator<KeyExtent> iter = lookupExtents(lookupRow).iterator();
                    while (iter.hasNext()) {
                        KeyExtent ke2 = iter.next();
                        if (inCache(ke2))
                            break;
                        updateCache(ke2);
                    }
                }
            }
        }
    }
}