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

Example 1 with TServerStatus

use of org.apache.accumulo.core.spi.balancer.data.TServerStatus 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 2 with TServerStatus

use of org.apache.accumulo.core.spi.balancer.data.TServerStatus 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;
}

Example 3 with TServerStatus

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

the class HostRegexTableLoadBalancerTest method testUnassignedWithNoDefaultPool.

@Test
public void testUnassignedWithNoDefaultPool() {
    init(DEFAULT_TABLE_PROPERTIES);
    Map<TabletId, TabletServerId> assignments = new HashMap<>();
    Map<TabletId, TabletServerId> unassigned = new HashMap<>();
    for (TabletId tabletId : tableTablets.get(BAR.getTableName())) {
        unassigned.put(tabletId, null);
    }
    SortedMap<TabletServerId, TServerStatus> current = createCurrent(15);
    // Remove the BAR tablet servers and default pool from current
    List<TabletServerId> removals = new ArrayList<>();
    for (Entry<TabletServerId, TServerStatus> e : current.entrySet()) {
        if (e.getKey().getHost().equals("192.168.0.6") || e.getKey().getHost().equals("192.168.0.7") || e.getKey().getHost().equals("192.168.0.8") || e.getKey().getHost().equals("192.168.0.9") || e.getKey().getHost().equals("192.168.0.10") || e.getKey().getHost().equals("192.168.0.11") || e.getKey().getHost().equals("192.168.0.12") || e.getKey().getHost().equals("192.168.0.13") || e.getKey().getHost().equals("192.168.0.14") || e.getKey().getHost().equals("192.168.0.15")) {
            removals.add(e.getKey());
        }
    }
    for (TabletServerId r : removals) {
        current.remove(r);
    }
    this.getAssignments(new AssignmentParamsImpl(Collections.unmodifiableSortedMap(current), Collections.unmodifiableMap(unassigned), assignments));
    assertEquals(unassigned.size(), assignments.size());
    // Ensure tablets are assigned in default pool
    for (Entry<TabletId, TabletServerId> e : assignments.entrySet()) {
        if (tabletInBounds(e.getKey(), e.getValue())) {
            fail("tablet unexpectedly in bounds: " + e.getKey() + " -> " + e.getValue().getHost());
        }
    }
}

Example 4 with TServerStatus

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

the class HostRegexTableLoadBalancerTest method testUnassignedWithNoTServers.

@Test
public void testUnassignedWithNoTServers() {
    init(DEFAULT_TABLE_PROPERTIES);
    Map<TabletId, TabletServerId> assignments = new HashMap<>();
    Map<TabletId, TabletServerId> unassigned = new HashMap<>();
    for (TabletId tabletId : tableTablets.get(BAR.getTableName())) {
        unassigned.put(tabletId, null);
    }
    SortedMap<TabletServerId, TServerStatus> current = createCurrent(15);
    // Remove the BAR tablet servers from current
    List<TabletServerId> removals = new ArrayList<>();
    for (Entry<TabletServerId, TServerStatus> e : current.entrySet()) {
        if (e.getKey().getHost().equals("192.168.0.6") || e.getKey().getHost().equals("192.168.0.7") || e.getKey().getHost().equals("192.168.0.8") || e.getKey().getHost().equals("192.168.0.9") || e.getKey().getHost().equals("192.168.0.10")) {
            removals.add(e.getKey());
        }
    }
    for (TabletServerId r : removals) {
        current.remove(r);
    }
    this.getAssignments(new AssignmentParamsImpl(Collections.unmodifiableSortedMap(current), Collections.unmodifiableMap(unassigned), assignments));
    assertEquals(unassigned.size(), assignments.size());
    // Ensure tablets are assigned in default pool
    for (Entry<TabletId, TabletServerId> e : assignments.entrySet()) {
        if (tabletInBounds(e.getKey(), e.getValue())) {
            fail("tablet unexpectedly in bounds: " + e.getKey() + " -> " + e.getValue().getHost());
        }
    }
}

Example 5 with TServerStatus

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

the class ChaoticLoadBalancer method balance.

@Override
public long balance(BalanceParameters params) {
    Map<TabletServerId, Long> numTablets = new HashMap<>();
    List<TabletServerId> underCapacityTServer = new ArrayList<>();
    if (!params.currentMigrations().isEmpty()) {
        outstandingMigrationsProblem.setMigrations(params.currentMigrations());
        problemReporter.reportProblem(outstandingMigrationsProblem);
        return 100;
    }
    problemReporter.clearProblemReportTimes();
    boolean moveMetadata = random.nextInt(4) == 0;
    long totalTablets = 0;
    for (Entry<TabletServerId, TServerStatus> e : params.currentStatus().entrySet()) {
        long tabletCount = 0;
        for (TableStatistics ti : e.getValue().getTableMap().values()) {
            tabletCount += ti.getTabletCount();
        }
        numTablets.put(e.getKey(), tabletCount);
        underCapacityTServer.add(e.getKey());
        totalTablets += tabletCount;
    }
    // totalTablets is fuzzy due to asynchronicity of the stats
    // *1.2 to handle fuzziness, and prevent locking for 'perfect' balancing scenarios
    long avg = (long) Math.ceil(((double) totalTablets) / params.currentStatus().size() * 1.2);
    for (Entry<TabletServerId, TServerStatus> e : params.currentStatus().entrySet()) {
        for (String tableId : e.getValue().getTableMap().keySet()) {
            TableId id = TableId.of(tableId);
            if (!moveMetadata && MetadataTable.ID.equals(id))
                continue;
            try {
                for (TabletStatistics ts : getOnlineTabletsForTable(e.getKey(), id)) {
                    int index = random.nextInt(underCapacityTServer.size());
                    TabletServerId dest = underCapacityTServer.get(index);
                    if (dest.equals(e.getKey()))
                        continue;
                    params.migrationsOut().add(new TabletMigration(ts.getTabletId(), e.getKey(), dest));
                    if (numTablets.put(dest, numTablets.get(dest) + 1) > avg)
                        underCapacityTServer.remove(index);
                    if (numTablets.put(e.getKey(), numTablets.get(e.getKey()) - 1) <= avg && !underCapacityTServer.contains(e.getKey()))
                        underCapacityTServer.add(e.getKey());
                    // option!
                    if (underCapacityTServer.isEmpty())
                        underCapacityTServer.addAll(numTablets.keySet());
                }
            } catch (AccumuloSecurityException e1) {
                // Shouldn't happen, but carry on if it does
                log.debug("Encountered AccumuloSecurityException.  This should not happen.  Carrying on anyway.", e1);
            } catch (AccumuloException e1) {
                // Shouldn't happen, but carry on if it does
                log.debug("Encountered AccumuloException.  This should not happen.  Carrying on anyway.", e1);
            }
        }
    }
    return 100;
}