コード例 #1
0
  private void sendPartitions() {
    ClusterNode oldestNode = this.oldestNode.get();

    try {
      sendLocalPartitions(oldestNode, exchId);
    } catch (ClusterTopologyCheckedException ignore) {
      if (log.isDebugEnabled())
        log.debug(
            "Oldest node left during partition exchange [nodeId="
                + oldestNode.id()
                + ", exchId="
                + exchId
                + ']');
    } catch (IgniteCheckedException e) {
      scheduleRecheck();

      U.error(
          log,
          "Failed to send local partitions to oldest node (will retry after timeout) [oldestNodeId="
              + oldestNode.id()
              + ", exchId="
              + exchId
              + ']',
          e);
    }
  }
コード例 #2
0
  /**
   * @param cctx Cache context.
   * @param busyLock Busy lock.
   * @param exchId Exchange ID.
   * @param reqs Cache change requests.
   */
  public GridDhtPartitionsExchangeFuture(
      GridCacheSharedContext cctx,
      ReadWriteLock busyLock,
      GridDhtPartitionExchangeId exchId,
      Collection<DynamicCacheChangeRequest> reqs) {
    assert busyLock != null;
    assert exchId != null;

    dummy = false;
    forcePreload = false;
    reassign = false;

    this.cctx = cctx;
    this.busyLock = busyLock;
    this.exchId = exchId;
    this.reqs = reqs;

    log = cctx.logger(getClass());

    initFut = new GridFutureAdapter<>();

    if (log.isDebugEnabled())
      log.debug(
          "Creating exchange future [localNode=" + cctx.localNodeId() + ", fut=" + this + ']');
  }
コード例 #3
0
  /**
   * @param nodes Nodes.
   * @param id ID.
   * @throws IgniteCheckedException If failed.
   */
  private void sendAllPartitions(
      Collection<? extends ClusterNode> nodes, GridDhtPartitionExchangeId id)
      throws IgniteCheckedException {
    GridDhtPartitionsFullMessage m =
        new GridDhtPartitionsFullMessage(id, lastVer.get(), id.topologyVersion());

    for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
      if (!cacheCtx.isLocal()) {
        AffinityTopologyVersion startTopVer = cacheCtx.startTopologyVersion();

        boolean ready = startTopVer == null || startTopVer.compareTo(id.topologyVersion()) <= 0;

        if (ready)
          m.addFullPartitionsMap(cacheCtx.cacheId(), cacheCtx.topology().partitionMap(true));
      }
    }

    // It is important that client topologies be added after contexts.
    for (GridClientPartitionTopology top : cctx.exchange().clientTopologies())
      m.addFullPartitionsMap(top.cacheId(), top.partitionMap(true));

    if (log.isDebugEnabled())
      log.debug(
          "Sending full partition map [nodeIds="
              + F.viewReadOnly(nodes, F.node2id())
              + ", exchId="
              + exchId
              + ", msg="
              + m
              + ']');

    cctx.io().safeSend(nodes, m, SYSTEM_POOL, null);
  }
コード例 #4
0
  /** @return {@code true} if entered to busy state. */
  private boolean enterBusy() {
    if (busyLock.readLock().tryLock()) return true;

    if (log.isDebugEnabled())
      log.debug("Failed to enter busy state (exchanger is stopping): " + this);

    return false;
  }
コード例 #5
0
  /** {@inheritDoc} */
  @Override
  public boolean onDone(AffinityTopologyVersion res, Throwable err) {
    Map<Integer, Boolean> m = null;

    for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
      if (cacheCtx.config().getTopologyValidator() != null && !CU.isSystemCache(cacheCtx.name())) {
        if (m == null) m = new HashMap<>();

        m.put(
            cacheCtx.cacheId(),
            cacheCtx.config().getTopologyValidator().validate(discoEvt.topologyNodes()));
      }
    }

    cacheValidRes = m != null ? m : Collections.<Integer, Boolean>emptyMap();

    cctx.cache().onExchangeDone(exchId.topologyVersion(), reqs, err);

    cctx.exchange().onExchangeDone(this, err);

    if (super.onDone(res, err) && !dummy && !forcePreload) {
      if (log.isDebugEnabled())
        log.debug(
            "Completed partition exchange [localNode="
                + cctx.localNodeId()
                + ", exchange= "
                + this
                + ']');

      initFut.onDone(err == null);

      GridTimeoutObject timeoutObj = this.timeoutObj;

      // Deschedule timeout object.
      if (timeoutObj != null) cctx.kernalContext().timeout().removeTimeoutObject(timeoutObj);

      if (exchId.isLeft()) {
        for (GridCacheContext cacheCtx : cctx.cacheContexts())
          cacheCtx.config().getAffinity().removeNode(exchId.nodeId());
      }

      return true;
    }

    return dummy;
  }
コード例 #6
0
  /**
   * @param node Node.
   * @param id ID.
   * @throws IgniteCheckedException If failed.
   */
  private void sendLocalPartitions(ClusterNode node, @Nullable GridDhtPartitionExchangeId id)
      throws IgniteCheckedException {
    GridDhtPartitionsSingleMessage m =
        new GridDhtPartitionsSingleMessage(
            id, cctx.kernalContext().clientNode(), cctx.versions().last());

    for (GridCacheContext cacheCtx : cctx.cacheContexts()) {
      if (!cacheCtx.isLocal())
        m.addLocalPartitionMap(cacheCtx.cacheId(), cacheCtx.topology().localPartitionMap());
    }

    if (log.isDebugEnabled())
      log.debug(
          "Sending local partitions [nodeId="
              + node.id()
              + ", exchId="
              + exchId
              + ", msg="
              + m
              + ']');

    cctx.io().send(node, m, SYSTEM_POOL);
  }
コード例 #7
0
  /**
   * @param nodeId Node ID.
   * @param retryCnt Number of retries.
   */
  private void sendAllPartitions(final UUID nodeId, final int retryCnt) {
    ClusterNode n = cctx.node(nodeId);

    try {
      if (n != null) sendAllPartitions(F.asList(n), exchId);
    } catch (IgniteCheckedException e) {
      if (e instanceof ClusterTopologyCheckedException || !cctx.discovery().alive(n)) {
        log.debug(
            "Failed to send full partition map to node, node left grid "
                + "[rmtNode="
                + nodeId
                + ", exchangeId="
                + exchId
                + ']');

        return;
      }

      if (retryCnt > 0) {
        long timeout = cctx.gridConfig().getNetworkSendRetryDelay();

        LT.error(
            log,
            e,
            "Failed to send full partition map to node (will retry after timeout) "
                + "[node="
                + nodeId
                + ", exchangeId="
                + exchId
                + ", timeout="
                + timeout
                + ']');

        cctx.time()
            .addTimeoutObject(
                new GridTimeoutObjectAdapter(timeout) {
                  @Override
                  public void onTimeout() {
                    sendAllPartitions(nodeId, retryCnt - 1);
                  }
                });
      } else
        U.error(
            log,
            "Failed to send full partition map [node=" + n + ", exchangeId=" + exchId + ']',
            e);
    }
  }
コード例 #8
0
  /**
   * @param cacheCtx Cache context.
   * @throws IgniteCheckedException If failed.
   */
  private void initTopology(GridCacheContext cacheCtx) throws IgniteCheckedException {
    if (stopping(cacheCtx.cacheId())) return;

    if (canCalculateAffinity(cacheCtx)) {
      if (log.isDebugEnabled())
        log.debug(
            "Will recalculate affinity [locNodeId="
                + cctx.localNodeId()
                + ", exchId="
                + exchId
                + ']');

      cacheCtx.affinity().calculateAffinity(exchId.topologyVersion(), discoEvt);
    } else {
      if (log.isDebugEnabled())
        log.debug(
            "Will request affinity from remote node [locNodeId="
                + cctx.localNodeId()
                + ", exchId="
                + exchId
                + ']');

      // Fetch affinity assignment from remote node.
      GridDhtAssignmentFetchFuture fetchFut =
          new GridDhtAssignmentFetchFuture(
              cacheCtx,
              exchId.topologyVersion(),
              CU.affinityNodes(cacheCtx, exchId.topologyVersion()));

      fetchFut.init();

      List<List<ClusterNode>> affAssignment = fetchFut.get();

      if (log.isDebugEnabled())
        log.debug(
            "Fetched affinity from remote node, initializing affinity assignment [locNodeId="
                + cctx.localNodeId()
                + ", topVer="
                + exchId.topologyVersion()
                + ']');

      if (affAssignment == null) {
        affAssignment = new ArrayList<>(cacheCtx.affinity().partitions());

        List<ClusterNode> empty = Collections.emptyList();

        for (int i = 0; i < cacheCtx.affinity().partitions(); i++) affAssignment.add(empty);
      }

      cacheCtx.affinity().initializeAffinity(exchId.topologyVersion(), affAssignment);
    }
  }
コード例 #9
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 #10
0
  /**
   * @param nodeId Sender node ID.
   * @param msg Full partition info.
   */
  public void onReceive(final UUID nodeId, final GridDhtPartitionsFullMessage msg) {
    assert msg != null;

    if (isDone()) {
      if (log.isDebugEnabled())
        log.debug("Received message for finished future [msg=" + msg + ", fut=" + this + ']');

      return;
    }

    if (log.isDebugEnabled())
      log.debug("Received full partition map from node [nodeId=" + nodeId + ", msg=" + msg + ']');

    assert exchId.topologyVersion().equals(msg.topologyVersion());

    initFut.listen(
        new CI1<IgniteInternalFuture<Boolean>>() {
          @Override
          public void apply(IgniteInternalFuture<Boolean> t) {
            ClusterNode curOldest = oldestNode.get();

            if (!nodeId.equals(curOldest.id())) {
              if (log.isDebugEnabled())
                log.debug(
                    "Received full partition map from unexpected node [oldest="
                        + curOldest.id()
                        + ", unexpectedNodeId="
                        + nodeId
                        + ']');

              ClusterNode snd = cctx.discovery().node(nodeId);

              if (snd == null) {
                if (log.isDebugEnabled())
                  log.debug(
                      "Sender node left grid, will ignore message from unexpected node [nodeId="
                          + nodeId
                          + ", exchId="
                          + msg.exchangeId()
                          + ']');

                return;
              }

              // Will process message later if sender node becomes oldest node.
              if (snd.order() > curOldest.order()) fullMsgs.put(nodeId, msg);

              return;
            }

            assert msg.exchangeId().equals(exchId);

            assert msg.lastVersion() != null;

            cctx.versions().onReceived(nodeId, msg.lastVersion());

            updatePartitionFullMap(msg);

            onDone(exchId.topologyVersion());
          }
        });
  }
コード例 #11
0
  /**
   * @param nodeId Sender node id.
   * @param msg Single partition info.
   */
  public void onReceive(final UUID nodeId, final GridDhtPartitionsSingleMessage msg) {
    assert msg != null;

    assert msg.exchangeId().equals(exchId);

    // Update last seen version.
    while (true) {
      GridCacheVersion old = lastVer.get();

      if (old == null || old.compareTo(msg.lastVersion()) < 0) {
        if (lastVer.compareAndSet(old, msg.lastVersion())) break;
      } else break;
    }

    if (isDone()) {
      if (log.isDebugEnabled())
        log.debug(
            "Received message for finished future (will reply only to sender) [msg="
                + msg
                + ", fut="
                + this
                + ']');

      sendAllPartitions(nodeId, cctx.gridConfig().getNetworkSendRetryCount());
    } else {
      initFut.listen(
          new CI1<IgniteInternalFuture<Boolean>>() {
            @Override
            public void apply(IgniteInternalFuture<Boolean> t) {
              try {
                if (!t.get()) // Just to check if there was an error.
                return;

                ClusterNode loc = cctx.localNode();

                singleMsgs.put(nodeId, msg);

                boolean match = true;

                // Check if oldest node has changed.
                if (!oldestNode.get().equals(loc)) {
                  match = false;

                  synchronized (mux) {
                    // Double check.
                    if (oldestNode.get().equals(loc)) match = true;
                  }
                }

                if (match) {
                  boolean allReceived;

                  synchronized (rcvdIds) {
                    if (rcvdIds.add(nodeId)) updatePartitionSingleMap(msg);

                    allReceived = allReceived();
                  }

                  // If got all replies, and initialization finished, and reply has not been sent
                  // yet.
                  if (allReceived && ready.get() && replied.compareAndSet(false, true)) {
                    spreadPartitions();

                    onDone(exchId.topologyVersion());
                  } else if (log.isDebugEnabled())
                    log.debug(
                        "Exchange future full map is not sent [allReceived="
                            + allReceived()
                            + ", ready="
                            + ready
                            + ", replied="
                            + replied.get()
                            + ", init="
                            + init.get()
                            + ", fut="
                            + this
                            + ']');
                }
              } catch (IgniteCheckedException e) {
                U.error(log, "Failed to initialize exchange future: " + this, e);
              }
            }
          });
    }
  }