/**
   * JUnit.
   *
   * @throws Exception If failed.
   */
  @SuppressWarnings({"TooBroadScope"})
  public void testRestarts() throws Exception {
    int duration = 60 * 1000;
    int qryThreadNum = 10;
    final long nodeLifeTime = 2 * 1000;
    final int logFreq = 20;

    final IgniteCache<Integer, Integer> cache = grid(0).cache(null);

    assert cache != null;

    for (int i = 0; i < KEY_CNT; i++) cache.put(i, i);

    assertEquals(KEY_CNT, cache.localSize());

    final AtomicInteger qryCnt = new AtomicInteger();

    final AtomicBoolean done = new AtomicBoolean();

    IgniteInternalFuture<?> fut1 =
        multithreadedAsync(
            new CAX() {
              @Override
              public void applyx() throws IgniteCheckedException {
                while (!done.get()) {
                  Collection<Cache.Entry<Integer, Integer>> res =
                      cache.query(new SqlQuery(Integer.class, "_val >= 0")).getAll();

                  assertFalse(res.isEmpty());

                  int c = qryCnt.incrementAndGet();

                  if (c % logFreq == 0) info("Executed queries: " + c);
                }
              }
            },
            qryThreadNum);

    final AtomicInteger restartCnt = new AtomicInteger();

    CollectingEventListener lsnr = new CollectingEventListener();

    for (int i = 0; i < GRID_CNT; i++)
      grid(i).events().localListen(lsnr, EventType.EVT_CACHE_REBALANCE_STOPPED);

    IgniteInternalFuture<?> fut2 =
        multithreadedAsync(
            new Callable<Object>() {
              @SuppressWarnings({"BusyWait"})
              @Override
              public Object call() throws Exception {
                while (!done.get()) {
                  int idx = GRID_CNT;

                  startGrid(idx);

                  Thread.sleep(nodeLifeTime);

                  stopGrid(idx);

                  int c = restartCnt.incrementAndGet();

                  if (c % logFreq == 0) info("Node restarts: " + c);
                }

                return true;
              }
            },
            1);

    Thread.sleep(duration);

    done.set(true);

    fut1.get();
    fut2.get();

    info("Awaiting rebalance events [restartCnt=" + restartCnt.get() + ']');

    boolean success = lsnr.awaitEvents(GRID_CNT * 2 * restartCnt.get(), 15000);

    for (int i = 0; i < GRID_CNT; i++)
      grid(i).events().stopLocalListen(lsnr, EventType.EVT_CACHE_REBALANCE_STOPPED);

    assert success;
  }
コード例 #2
0
  /**
   * Starts activity.
   *
   * @throws IgniteInterruptedCheckedException If interrupted.
   */
  public void init() throws IgniteInterruptedCheckedException {
    if (isDone()) return;

    if (init.compareAndSet(false, true)) {
      if (isDone()) return;

      try {
        // Wait for event to occur to make sure that discovery
        // will return corresponding nodes.
        U.await(evtLatch);

        assert discoEvt != null : this;
        assert !dummy && !forcePreload : this;

        ClusterNode oldest = CU.oldestAliveCacheServerNode(cctx, exchId.topologyVersion());

        oldestNode.set(oldest);

        startCaches();

        // True if client node joined or failed.
        boolean clientNodeEvt;

        if (F.isEmpty(reqs)) {
          int type = discoEvt.type();

          assert type == EVT_NODE_JOINED || type == EVT_NODE_LEFT || type == EVT_NODE_FAILED
              : discoEvt;

          clientNodeEvt = CU.clientNode(discoEvt.eventNode());
        } else {
          assert discoEvt.type() == EVT_DISCOVERY_CUSTOM_EVT : discoEvt;

          boolean clientOnlyStart = true;

          for (DynamicCacheChangeRequest req : reqs) {
            if (!req.clientStartOnly()) {
              clientOnlyStart = false;

              break;
            }
          }

          clientNodeEvt = clientOnlyStart;
        }

        if (clientNodeEvt) {
          ClusterNode node = discoEvt.eventNode();

          // Client need to initialize affinity for local join event or for stated client caches.
          if (!node.isLocal()) {
            for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
              if (cacheCtx.isLocal()) continue;

              GridDhtPartitionTopology top = cacheCtx.topology();

              top.updateTopologyVersion(exchId, this, -1, stopping(cacheCtx.cacheId()));

              if (cacheCtx.affinity().affinityTopologyVersion() == AffinityTopologyVersion.NONE) {
                initTopology(cacheCtx);

                top.beforeExchange(this);
              } else
                cacheCtx.affinity().clientEventTopologyChange(discoEvt, exchId.topologyVersion());
            }

            if (exchId.isLeft())
              cctx.mvcc().removeExplicitNodeLocks(exchId.nodeId(), exchId.topologyVersion());

            onDone(exchId.topologyVersion());

            skipPreload = cctx.kernalContext().clientNode();

            return;
          }
        }

        if (cctx.kernalContext().clientNode()) {
          skipPreload = true;

          for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
            if (cacheCtx.isLocal()) continue;

            GridDhtPartitionTopology top = cacheCtx.topology();

            top.updateTopologyVersion(exchId, this, -1, stopping(cacheCtx.cacheId()));
          }

          for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
            if (cacheCtx.isLocal()) continue;

            initTopology(cacheCtx);
          }

          if (oldestNode.get() != null) {
            rmtNodes =
                new ConcurrentLinkedQueue<>(
                    CU.aliveRemoteServerNodesWithCaches(cctx, exchId.topologyVersion()));

            rmtIds = Collections.unmodifiableSet(new HashSet<>(F.nodeIds(rmtNodes)));

            ready.set(true);

            initFut.onDone(true);

            if (log.isDebugEnabled()) log.debug("Initialized future: " + this);

            sendPartitions();
          } else onDone(exchId.topologyVersion());

          return;
        }

        assert oldestNode.get() != null;

        for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
          if (isCacheAdded(cacheCtx.cacheId(), exchId.topologyVersion())) {
            if (cacheCtx
                .discovery()
                .cacheAffinityNodes(cacheCtx.name(), topologyVersion())
                .isEmpty())
              U.quietAndWarn(log, "No server nodes found for cache client: " + cacheCtx.namex());
          }

          cacheCtx.preloader().onExchangeFutureAdded();
        }

        List<String> cachesWithoutNodes = null;

        if (exchId.isLeft()) {
          for (String name : cctx.cache().cacheNames()) {
            if (cctx.discovery().cacheAffinityNodes(name, topologyVersion()).isEmpty()) {
              if (cachesWithoutNodes == null) cachesWithoutNodes = new ArrayList<>();

              cachesWithoutNodes.add(name);

              // Fire event even if there is no client cache started.
              if (cctx.gridEvents().isRecordable(EventType.EVT_CACHE_NODES_LEFT)) {
                Event evt =
                    new CacheEvent(
                        name,
                        cctx.localNode(),
                        cctx.localNode(),
                        "All server nodes have left the cluster.",
                        EventType.EVT_CACHE_NODES_LEFT,
                        0,
                        false,
                        null,
                        null,
                        null,
                        null,
                        false,
                        null,
                        false,
                        null,
                        null,
                        null);

                cctx.gridEvents().record(evt);
              }
            }
          }
        }

        if (cachesWithoutNodes != null) {
          StringBuilder sb =
              new StringBuilder(
                  "All server nodes for the following caches have left the cluster: ");

          for (int i = 0; i < cachesWithoutNodes.size(); i++) {
            String cache = cachesWithoutNodes.get(i);

            sb.append('\'').append(cache).append('\'');

            if (i != cachesWithoutNodes.size() - 1) sb.append(", ");
          }

          U.quietAndWarn(log, sb.toString());

          U.quietAndWarn(log, "Must have server nodes for caches to operate.");
        }

        assert discoEvt != null;

        assert exchId.nodeId().equals(discoEvt.eventNode().id());

        for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
          GridClientPartitionTopology clientTop =
              cctx.exchange().clearClientTopology(cacheCtx.cacheId());

          long updSeq = clientTop == null ? -1 : clientTop.lastUpdateSequence();

          // Update before waiting for locks.
          if (!cacheCtx.isLocal())
            cacheCtx
                .topology()
                .updateTopologyVersion(exchId, this, updSeq, stopping(cacheCtx.cacheId()));
        }

        // Grab all alive remote nodes with order of equal or less than last joined node.
        rmtNodes =
            new ConcurrentLinkedQueue<>(
                CU.aliveRemoteServerNodesWithCaches(cctx, exchId.topologyVersion()));

        rmtIds = Collections.unmodifiableSet(new HashSet<>(F.nodeIds(rmtNodes)));

        for (Map.Entry<UUID, GridDhtPartitionsSingleMessage> m : singleMsgs.entrySet())
          // If received any messages, process them.
          onReceive(m.getKey(), m.getValue());

        for (Map.Entry<UUID, GridDhtPartitionsFullMessage> m : fullMsgs.entrySet())
          // If received any messages, process them.
          onReceive(m.getKey(), m.getValue());

        AffinityTopologyVersion topVer = exchId.topologyVersion();

        for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
          if (cacheCtx.isLocal()) continue;

          // Must initialize topology after we get discovery event.
          initTopology(cacheCtx);

          cacheCtx.preloader().updateLastExchangeFuture(this);
        }

        IgniteInternalFuture<?> partReleaseFut = cctx.partitionReleaseFuture(topVer);

        // Assign to class variable so it will be included into toString() method.
        this.partReleaseFut = partReleaseFut;

        if (log.isDebugEnabled()) log.debug("Before waiting for partition release future: " + this);

        while (true) {
          try {
            partReleaseFut.get(2 * cctx.gridConfig().getNetworkTimeout(), TimeUnit.MILLISECONDS);

            break;
          } catch (IgniteFutureTimeoutCheckedException ignored) {
            // Print pending transactions and locks that might have led to hang.
            dumpPendingObjects();
          }
        }

        if (log.isDebugEnabled()) log.debug("After waiting for partition release future: " + this);

        if (!F.isEmpty(reqs)) blockGateways();

        if (exchId.isLeft())
          cctx.mvcc().removeExplicitNodeLocks(exchId.nodeId(), exchId.topologyVersion());

        IgniteInternalFuture<?> locksFut = cctx.mvcc().finishLocks(exchId.topologyVersion());

        while (true) {
          try {
            locksFut.get(2 * cctx.gridConfig().getNetworkTimeout(), TimeUnit.MILLISECONDS);

            break;
          } catch (IgniteFutureTimeoutCheckedException ignored) {
            U.warn(
                log,
                "Failed to wait for locks release future. "
                    + "Dumping pending objects that might be the cause: "
                    + cctx.localNodeId());

            U.warn(log, "Locked entries:");

            Map<IgniteTxKey, Collection<GridCacheMvccCandidate>> locks =
                cctx.mvcc().unfinishedLocks(exchId.topologyVersion());

            for (Map.Entry<IgniteTxKey, Collection<GridCacheMvccCandidate>> e : locks.entrySet())
              U.warn(log, "Locked entry [key=" + e.getKey() + ", mvcc=" + e.getValue() + ']');
          }
        }

        for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
          if (cacheCtx.isLocal()) continue;

          // Notify replication manager.
          GridCacheContext drCacheCtx =
              cacheCtx.isNear() ? cacheCtx.near().dht().context() : cacheCtx;

          if (drCacheCtx.isDrEnabled()) drCacheCtx.dr().beforeExchange(topVer, exchId.isLeft());

          // Partition release future is done so we can flush the write-behind store.
          cacheCtx.store().forceFlush();

          // Process queued undeploys prior to sending/spreading map.
          cacheCtx.preloader().unwindUndeploys();

          GridDhtPartitionTopology top = cacheCtx.topology();

          assert topVer.equals(top.topologyVersion())
              : "Topology version is updated only in this class instances inside single ExchangeWorker thread.";

          top.beforeExchange(this);
        }

        for (GridClientPartitionTopology top : cctx.exchange().clientTopologies()) {
          top.updateTopologyVersion(exchId, this, -1, stopping(top.cacheId()));

          top.beforeExchange(this);
        }
      } catch (IgniteInterruptedCheckedException e) {
        onDone(e);

        throw e;
      } catch (Throwable e) {
        U.error(
            log,
            "Failed to reinitialize local partitions (preloading will be stopped): " + exchId,
            e);

        onDone(e);

        if (e instanceof Error) throw (Error) e;

        return;
      }

      if (F.isEmpty(rmtIds)) {
        onDone(exchId.topologyVersion());

        return;
      }

      ready.set(true);

      initFut.onDone(true);

      if (log.isDebugEnabled()) log.debug("Initialized future: " + this);

      // If this node is not oldest.
      if (!oldestNode.get().id().equals(cctx.localNodeId())) sendPartitions();
      else {
        boolean allReceived = allReceived();

        if (allReceived && replied.compareAndSet(false, true)) {
          if (spreadPartitions()) onDone(exchId.topologyVersion());
        }
      }

      scheduleRecheck();
    } else assert false : "Skipped init future: " + this;
  }
コード例 #3
0
  /**
   * @param cacheMode Cache mode.
   * @param sameAff If {@code false} uses different number of partitions for caches.
   * @param concurrency Transaction concurrency.
   * @param isolation Transaction isolation.
   * @throws Exception If failed.
   */
  private void crossCacheTxFailover(
      CacheMode cacheMode,
      boolean sameAff,
      final TransactionConcurrency concurrency,
      final TransactionIsolation isolation)
      throws Exception {
    IgniteKernal ignite0 = (IgniteKernal) ignite(0);

    final AtomicBoolean stop = new AtomicBoolean();

    try {
      ignite0.createCache(cacheConfiguration(CACHE1, cacheMode, 256));
      ignite0.createCache(cacheConfiguration(CACHE2, cacheMode, sameAff ? 256 : 128));

      final AtomicInteger threadIdx = new AtomicInteger();

      IgniteInternalFuture<?> fut =
          GridTestUtils.runMultiThreadedAsync(
              new Callable<Void>() {
                @Override
                public Void call() throws Exception {
                  int idx = threadIdx.getAndIncrement();

                  Ignite ignite = ignite(idx % GRID_CNT);

                  log.info(
                      "Started update thread [node="
                          + ignite.name()
                          + ", client="
                          + ignite.configuration().isClientMode()
                          + ']');

                  IgniteCache<TestKey, TestValue> cache1 = ignite.cache(CACHE1);
                  IgniteCache<TestKey, TestValue> cache2 = ignite.cache(CACHE2);

                  assertNotSame(cache1, cache2);

                  IgniteTransactions txs = ignite.transactions();

                  ThreadLocalRandom rnd = ThreadLocalRandom.current();

                  long iter = 0;

                  while (!stop.get()) {
                    boolean sameKey = rnd.nextBoolean();

                    try {
                      try (Transaction tx = txs.txStart(concurrency, isolation)) {
                        if (sameKey) {
                          TestKey key = new TestKey(rnd.nextLong(KEY_RANGE));

                          cacheOperation(rnd, cache1, key);
                          cacheOperation(rnd, cache2, key);
                        } else {
                          TestKey key1 = new TestKey(rnd.nextLong(KEY_RANGE));
                          TestKey key2 = new TestKey(key1.key() + 1);

                          cacheOperation(rnd, cache1, key1);
                          cacheOperation(rnd, cache2, key2);
                        }

                        tx.commit();
                      }
                    } catch (CacheException | IgniteException e) {
                      log.info("Update error: " + e);
                    }

                    if (iter++ % 500 == 0) log.info("Iteration: " + iter);
                  }

                  return null;
                }

                /**
                 * @param rnd Random.
                 * @param cache Cache.
                 * @param key Key.
                 */
                private void cacheOperation(
                    ThreadLocalRandom rnd, IgniteCache<TestKey, TestValue> cache, TestKey key) {
                  switch (rnd.nextInt(4)) {
                    case 0:
                      cache.put(key, new TestValue(rnd.nextLong()));

                      break;

                    case 1:
                      cache.remove(key);

                      break;

                    case 2:
                      cache.invoke(key, new TestEntryProcessor(rnd.nextBoolean() ? 1L : null));

                      break;

                    case 3:
                      cache.get(key);

                      break;

                    default:
                      assert false;
                  }
                }
              },
              10,
              "tx-thread");

      long stopTime = System.currentTimeMillis() + 3 * 60_000;

      long topVer = ignite0.cluster().topologyVersion();

      boolean failed = false;

      while (System.currentTimeMillis() < stopTime) {
        log.info("Start node.");

        IgniteKernal ignite = (IgniteKernal) startGrid(GRID_CNT);

        assertFalse(ignite.configuration().isClientMode());

        topVer++;

        IgniteInternalFuture<?> affFut =
            ignite
                .context()
                .cache()
                .context()
                .exchange()
                .affinityReadyFuture(new AffinityTopologyVersion(topVer));

        try {
          if (affFut != null) affFut.get(30_000);
        } catch (IgniteFutureTimeoutCheckedException e) {
          log.error("Failed to wait for affinity future after start: " + topVer);

          failed = true;

          break;
        }

        Thread.sleep(500);

        log.info("Stop node.");

        stopGrid(GRID_CNT);

        topVer++;

        affFut =
            ignite0
                .context()
                .cache()
                .context()
                .exchange()
                .affinityReadyFuture(new AffinityTopologyVersion(topVer));

        try {
          if (affFut != null) affFut.get(30_000);
        } catch (IgniteFutureTimeoutCheckedException e) {
          log.error("Failed to wait for affinity future after stop: " + topVer);

          failed = true;

          break;
        }
      }

      stop.set(true);

      fut.get();

      assertFalse("Test failed, see log for details.", failed);
    } finally {
      stop.set(true);

      ignite0.destroyCache(CACHE1);
      ignite0.destroyCache(CACHE2);

      awaitPartitionMapExchange();
    }
  }