Examples with TabletServerId org.apache.accumulo.core.spi.balancer.data.TabletServerId used on opensource projects

Example 1 with TabletServerId

use of org.apache.accumulo.core.spi.balancer.data.TabletServerId in project accumulo by apache.

the class GroupBalancer method populateMigrations.

private void populateMigrations(Set<TabletServerId> current, List<TabletMigration> migrationsOut, Moves moves) {
    if (moves.size() == 0) {
        return;
    }
    Function<TabletId, String> partitioner = getPartitioner();
    for (var tablet : getLocationProvider().entrySet()) {
        String group = partitioner.apply(tablet.getKey());
        var loc = tablet.getValue();
        if (loc == null || !current.contains(loc)) {
            migrationsOut.clear();
            return;
        }
        TabletServerId dest = moves.removeMove(loc, group);
        if (dest != null) {
            migrationsOut.add(new TabletMigration(tablet.getKey(), loc, dest));
            if (moves.size() == 0) {
                break;
            }
        }
    }
}

Example 2 with TabletServerId

use of org.apache.accumulo.core.spi.balancer.data.TabletServerId in project accumulo by apache.

the class GroupBalancer method balance.

@Override
public long balance(BalanceParameters params) {
    if (!shouldBalance(params.currentStatus(), params.currentMigrations())) {
        return 5000;
    }
    if (System.currentTimeMillis() - lastRun < getWaitTime()) {
        return 5000;
    }
    MapCounter<String> groupCounts = new MapCounter<>();
    Map<TabletServerId, TserverGroupInfo> tservers = new HashMap<>();
    for (TabletServerId tsi : params.currentStatus().keySet()) {
        tservers.put(tsi, new TserverGroupInfo(tsi));
    }
    Function<TabletId, String> partitioner = getPartitioner();
    // collect stats about current state
    for (var tablet : getLocationProvider().entrySet()) {
        String group = partitioner.apply(tablet.getKey());
        var loc = tablet.getValue();
        if (loc == null || !tservers.containsKey(loc)) {
            return 5000;
        }
        groupCounts.increment(group, 1);
        TserverGroupInfo tgi = tservers.get(loc);
        tgi.addGroup(group);
    }
    Map<String, Integer> expectedCounts = new HashMap<>();
    int totalExtra = 0;
    for (String group : groupCounts.keySet()) {
        int groupCount = groupCounts.getInt(group);
        totalExtra += groupCount % params.currentStatus().size();
        expectedCounts.put(group, (groupCount / params.currentStatus().size()));
    }
    // The number of extra tablets from all groups that each tserver must have.
    int expectedExtra = totalExtra / params.currentStatus().size();
    int maxExtraGroups = expectedExtra + 1;
    expectedCounts = Collections.unmodifiableMap(expectedCounts);
    tservers = Collections.unmodifiableMap(tservers);
    for (TserverGroupInfo tgi : tservers.values()) {
        tgi.finishedAdding(expectedCounts);
    }
    Moves moves = new Moves();
    // The order of the following steps is important, because as ordered each step should not move
    // any tablets moved by a previous step.
    balanceExpected(tservers, moves);
    if (moves.size() < getMaxMigrations()) {
        balanceExtraExpected(tservers, expectedExtra, moves);
        if (moves.size() < getMaxMigrations()) {
            boolean cont = balanceExtraMultiple(tservers, maxExtraGroups, moves);
            if (cont && moves.size() < getMaxMigrations()) {
                balanceExtraExtra(tservers, maxExtraGroups, moves);
            }
        }
    }
    populateMigrations(tservers.keySet(), params.migrationsOut(), moves);
    lastRun = System.currentTimeMillis();
    return 5000;
}

Example 3 with TabletServerId

use of org.apache.accumulo.core.spi.balancer.data.TabletServerId in project accumulo by apache.

the class TableLoadBalancer method getAssignments.

@Override
public void getAssignments(AssignmentParameters params) {
    // separate the unassigned into tables
    Map<TableId, Map<TabletId, TabletServerId>> groupedUnassigned = new HashMap<>();
    params.unassignedTablets().forEach((tid, lastTserver) -> groupedUnassigned.computeIfAbsent(tid.getTable(), k -> new HashMap<>()).put(tid, lastTserver));
    for (Entry<TableId, Map<TabletId, TabletServerId>> e : groupedUnassigned.entrySet()) {
        Map<TabletId, TabletServerId> newAssignments = new HashMap<>();
        getBalancerForTable(e.getKey()).getAssignments(new AssignmentParamsImpl(params.currentStatus(), e.getValue(), newAssignments));
        newAssignments.forEach(params::addAssignment);
    }
}

Example 4 with TabletServerId

use of org.apache.accumulo.core.spi.balancer.data.TabletServerId in project accumulo by apache.

the class Manager method gatherTableInformation.

private SortedMap<TServerInstance, TabletServerStatus> gatherTableInformation(Set<TServerInstance> currentServers, SortedMap<TabletServerId, TServerStatus> balancerMap) {
    final long rpcTimeout = getConfiguration().getTimeInMillis(Property.GENERAL_RPC_TIMEOUT);
    int threads = getConfiguration().getCount(Property.MANAGER_STATUS_THREAD_POOL_SIZE);
    ExecutorService tp = ThreadPools.createExecutorService(getConfiguration(), Property.MANAGER_STATUS_THREAD_POOL_SIZE, false);
    long start = System.currentTimeMillis();
    final SortedMap<TServerInstance, TabletServerStatus> result = new ConcurrentSkipListMap<>();
    final RateLimiter shutdownServerRateLimiter = RateLimiter.create(MAX_SHUTDOWNS_PER_SEC);
    for (TServerInstance serverInstance : currentServers) {
        final TServerInstance server = serverInstance;
        if (threads == 0) {
            // Since an unbounded thread pool is being used, rate limit how fast task are added to the
            // executor. This prevents the threads from growing large unless there are lots of
            // unresponsive tservers.
            sleepUninterruptibly(Math.max(1, rpcTimeout / 120_000), TimeUnit.MILLISECONDS);
        }
        tp.submit(() -> {
            try {
                Thread t = Thread.currentThread();
                String oldName = t.getName();
                try {
                    String message = "Getting status from " + server;
                    t.setName(message);
                    long startForServer = System.currentTimeMillis();
                    log.trace(message);
                    TServerConnection connection1 = tserverSet.getConnection(server);
                    if (connection1 == null) {
                        throw new IOException("No connection to " + server);
                    }
                    TabletServerStatus status = connection1.getTableMap(false);
                    result.put(server, status);
                    long duration = System.currentTimeMillis() - startForServer;
                    log.trace("Got status from {} in {} ms", server, duration);
                } finally {
                    t.setName(oldName);
                }
            } catch (Exception ex) {
                log.error("unable to get tablet server status {} {}", server, ex.toString());
                log.debug("unable to get tablet server status {}", server, ex);
                // MAX_BAD_STATUS_COUNT times
                if (badServers.computeIfAbsent(server, k -> new AtomicInteger(0)).incrementAndGet() > MAX_BAD_STATUS_COUNT) {
                    if (shutdownServerRateLimiter.tryAcquire()) {
                        log.warn("attempting to stop {}", server);
                        try {
                            TServerConnection connection2 = tserverSet.getConnection(server);
                            if (connection2 != null) {
                                connection2.halt(managerLock);
                            }
                        } catch (TTransportException e1) {
                        // ignore: it's probably down
                        } catch (Exception e2) {
                            log.info("error talking to troublesome tablet server", e2);
                        }
                    } else {
                        log.warn("Unable to shutdown {} as over the shutdown limit of {} per minute", server, MAX_SHUTDOWNS_PER_SEC * 60);
                    }
                    badServers.remove(server);
                }
            }
        });
    }
    tp.shutdown();
    try {
        tp.awaitTermination(Math.max(10000, rpcTimeout / 3), TimeUnit.MILLISECONDS);
    } catch (InterruptedException e) {
        log.debug("Interrupted while fetching status");
    }
    tp.shutdownNow();
    // Threads may still modify map after shutdownNow is called, so create an immutable snapshot.
    SortedMap<TServerInstance, TabletServerStatus> info = ImmutableSortedMap.copyOf(result);
    tserverStatus.forEach((tsi, status) -> balancerMap.put(new TabletServerIdImpl(tsi), TServerStatusImpl.fromThrift(status)));
    synchronized (badServers) {
        badServers.keySet().retainAll(currentServers);
        badServers.keySet().removeAll(info.keySet());
    }
    log.debug(String.format("Finished gathering information from %d of %d servers in %.2f seconds", info.size(), currentServers.size(), (System.currentTimeMillis() - start) / 1000.));
    return info;
}

Example 5 with TabletServerId

use of org.apache.accumulo.core.spi.balancer.data.TabletServerId in project accumulo by apache.

the class BaseHostRegexTableLoadBalancerTest method createCurrent.

protected SortedMap<TabletServerId, TServerStatus> createCurrent(int numTservers) {
    String base = "192.168.0.";
    TreeMap<TabletServerId, TServerStatus> current = new TreeMap<>();
    for (int i = 1; i <= numTservers; i++) {
        TServerStatusImpl status = new TServerStatusImpl(new org.apache.accumulo.core.master.thrift.TabletServerStatus());
        Map<String, TableStatistics> tableMap = new HashMap<>();
        tableMap.put(FOO.getId().canonical(), new TableStatisticsImpl(new TableInfo()));
        tableMap.put(BAR.getId().canonical(), new TableStatisticsImpl(new TableInfo()));
        tableMap.put(BAZ.getId().canonical(), new TableStatisticsImpl(new TableInfo()));
        status.setTableMap(tableMap);
        current.put(new TabletServerIdImpl(base + i, 9997, Integer.toHexString(1)), status);
    }
    // now put all of the tablets on one server
    for (Map.Entry<String, TabletServerId> entry : initialTableLocation.entrySet()) {
        TServerStatus status = current.get(entry.getValue());
        if (status != null) {
            TableId tableId = environment.getTableIdMap().get(entry.getKey());
            ((TableStatisticsImpl) status.getTableMap().get(tableId.canonical())).setOnlineTabletCount(5);
        }
    }
    return current;
}