Exemplo n.º 1
0
  /** @param ignite Node. */
  public void dumpCacheDebugInfo(Ignite ignite) {
    GridKernalContext ctx = ((IgniteKernal) ignite).context();

    log.error(
        "Cache information update [node="
            + ignite.name()
            + ", client="
            + ignite.configuration().isClientMode()
            + ']');

    GridCacheSharedContext cctx = ctx.cache().context();

    log.error("Pending transactions:");

    for (IgniteInternalTx tx : cctx.tm().activeTransactions()) log.error(">>> " + tx);

    log.error("Pending explicit locks:");

    for (GridCacheExplicitLockSpan lockSpan : cctx.mvcc().activeExplicitLocks())
      log.error(">>> " + lockSpan);

    log.error("Pending cache futures:");

    for (GridCacheFuture<?> fut : cctx.mvcc().activeFutures()) log.error(">>> " + fut);

    log.error("Pending atomic cache futures:");

    for (GridCacheFuture<?> fut : cctx.mvcc().atomicFutures()) log.error(">>> " + fut);
  }
Exemplo n.º 2
0
  /** {@inheritDoc} */
  @Override
  public boolean onDone(IgniteInternalTx tx, Throwable err) {
    if (initialized() || err != null) {
      if (this.tx.onePhaseCommit() && (this.tx.state() == COMMITTING))
        this.tx.tmFinish(err == null);

      Throwable e = this.err;

      if (e == null && commit) e = this.tx.commitError();

      Throwable finishErr = e != null ? e : err;

      if (super.onDone(tx, finishErr)) {
        if (finishErr == null) finishErr = this.tx.commitError();

        if (this.tx.syncMode() != PRIMARY_SYNC) this.tx.sendFinishReply(commit, finishErr);

        // Don't forget to clean up.
        cctx.mvcc().removeFuture(futId);

        return true;
      }
    }

    return false;
  }
  private void dumpPendingObjects() {
    U.warn(
        log,
        "Failed to wait for partition release future. Dumping pending objects that might be the cause: "
            + cctx.localNodeId());

    U.warn(log, "Pending transactions:");

    for (IgniteInternalTx tx : cctx.tm().activeTransactions()) U.warn(log, ">>> " + tx);

    U.warn(log, "Pending explicit locks:");

    for (GridCacheExplicitLockSpan lockSpan : cctx.mvcc().activeExplicitLocks())
      U.warn(log, ">>> " + lockSpan);

    U.warn(log, "Pending cache futures:");

    for (GridCacheFuture<?> fut : cctx.mvcc().activeFutures()) U.warn(log, ">>> " + fut);

    U.warn(log, "Pending atomic cache futures:");

    for (GridCacheFuture<?> fut : cctx.mvcc().atomicFutures()) U.warn(log, ">>> " + fut);
  }
  /**
   * 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;
  }
  /** {@inheritDoc} */
  @Override
  public boolean onDone(IgniteInternalTx tx0, Throwable err) {
    if (isDone()) return false;

    synchronized (this) {
      if (isDone()) return false;

      if (err != null) {
        tx.commitError(err);

        boolean marked = tx.setRollbackOnly();

        if (err instanceof IgniteTxRollbackCheckedException) {
          if (marked) {
            try {
              tx.rollback();
            } catch (IgniteCheckedException ex) {
              U.error(log, "Failed to automatically rollback transaction: " + tx, ex);
            }
          }
        } else if (tx.implicit()
            && tx.isSystemInvalidate()) { // Finish implicit transaction on heuristic error.
          try {
            tx.close();
          } catch (IgniteCheckedException ex) {
            U.error(log, "Failed to invalidate transaction: " + tx, ex);
          }
        }
      }

      if (initialized() || err != null) {
        if (tx.needCheckBackup()) {
          assert tx.onePhaseCommit();

          if (err != null)
            err = new TransactionRollbackException("Failed to commit transaction.", err);

          try {
            tx.finish(err == null);
          } catch (IgniteCheckedException e) {
            if (err != null) err.addSuppressed(e);
            else err = e;
          }
        }

        if (tx.onePhaseCommit()) {
          boolean commit = this.commit && err == null;

          finishOnePhase(commit);

          tx.tmFinish(commit);
        }

        if (super.onDone(tx0, err)) {
          if (error() instanceof IgniteTxHeuristicCheckedException) {
            AffinityTopologyVersion topVer = tx.topologyVersion();

            for (IgniteTxEntry e : tx.writeMap().values()) {
              GridCacheContext cacheCtx = e.context();

              try {
                if (e.op() != NOOP && !cacheCtx.affinity().localNode(e.key(), topVer)) {
                  GridCacheEntryEx entry = cacheCtx.cache().peekEx(e.key());

                  if (entry != null) entry.invalidate(null, tx.xidVersion());
                }
              } catch (Throwable t) {
                U.error(log, "Failed to invalidate entry.", t);

                if (t instanceof Error) throw (Error) t;
              }
            }
          }

          // Don't forget to clean up.
          cctx.mvcc().removeFuture(futId);

          return true;
        }
      }
    }

    return false;
  }