Examples with GridDhtPartitionTopology org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology used on opensource projects

Example 1 with GridDhtPartitionTopology

use of org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology in project ignite by apache.

the class GridDhtPreloader method assign.

/** {@inheritDoc} */
@Override
public GridDhtPreloaderAssignments assign(GridDhtPartitionsExchangeFuture exchFut) {
    // No assignments for disabled preloader.
    GridDhtPartitionTopology top = cctx.dht().topology();
    if (!cctx.rebalanceEnabled() || !cctx.shared().kernalContext().state().active())
        return new GridDhtPreloaderAssignments(exchFut, top.topologyVersion());
    int partCnt = cctx.affinity().partitions();
    assert exchFut.forcePreload() || exchFut.dummyReassign() || exchFut.exchangeId().topologyVersion().equals(top.topologyVersion()) : "Topology version mismatch [exchId=" + exchFut.exchangeId() + ", cache=" + cctx.name() + ", topVer=" + top.topologyVersion() + ']';
    GridDhtPreloaderAssignments assigns = new GridDhtPreloaderAssignments(exchFut, top.topologyVersion());
    AffinityTopologyVersion topVer = assigns.topologyVersion();
    for (int p = 0; p < partCnt; p++) {
        if (cctx.shared().exchange().hasPendingExchange()) {
            if (log.isDebugEnabled())
                log.debug("Skipping assignments creation, exchange worker has pending assignments: " + exchFut.exchangeId());
            assigns.cancelled(true);
            return assigns;
        }
        // If partition belongs to local node.
        if (cctx.affinity().partitionLocalNode(p, topVer)) {
            GridDhtLocalPartition part = top.localPartition(p, topVer, true);
            assert part != null;
            assert part.id() == p;
            if (part.state() != MOVING) {
                if (log.isDebugEnabled())
                    log.debug("Skipping partition assignment (state is not MOVING): " + part);
                // For.
                continue;
            }
            Collection<ClusterNode> picked = pickedOwners(p, topVer);
            if (picked.isEmpty()) {
                top.own(part);
                if (cctx.events().isRecordable(EVT_CACHE_REBALANCE_PART_DATA_LOST)) {
                    DiscoveryEvent discoEvt = exchFut.discoveryEvent();
                    cctx.events().addPreloadEvent(p, EVT_CACHE_REBALANCE_PART_DATA_LOST, discoEvt.eventNode(), discoEvt.type(), discoEvt.timestamp());
                }
                if (log.isDebugEnabled())
                    log.debug("Owning partition as there are no other owners: " + part);
            } else {
                ClusterNode n = F.rand(picked);
                GridDhtPartitionDemandMessage msg = assigns.get(n);
                if (msg == null) {
                    assigns.put(n, msg = new GridDhtPartitionDemandMessage(top.updateSequence(), exchFut.exchangeId().topologyVersion(), cctx.cacheId()));
                }
                msg.addPartition(p);
            }
        }
    }
    return assigns;
}

Example 2 with GridDhtPartitionTopology

use of org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology in project ignite by apache.

the class IgniteTxHandler method prepareNearTx.

/**
 * @param nearNode Node that initiated transaction.
 * @param req Near prepare request.
 * @return Prepare future or {@code null} if need retry operation.
 */
@Nullable
private IgniteInternalFuture<GridNearTxPrepareResponse> prepareNearTx(final ClusterNode nearNode, final GridNearTxPrepareRequest req) {
    IgniteTxEntry firstEntry;
    try {
        IgniteTxEntry firstWrite = unmarshal(req.writes());
        IgniteTxEntry firstRead = unmarshal(req.reads());
        firstEntry = firstWrite != null ? firstWrite : firstRead;
    } catch (IgniteCheckedException e) {
        return new GridFinishedFuture<>(e);
    }
    assert firstEntry != null : req;
    GridDhtTxLocal tx = null;
    GridCacheVersion mappedVer = ctx.tm().mappedVersion(req.version());
    if (mappedVer != null) {
        tx = ctx.tm().tx(mappedVer);
        if (tx == null)
            U.warn(log, "Missing local transaction for mapped near version [nearVer=" + req.version() + ", mappedVer=" + mappedVer + ']');
        else {
            if (req.concurrency() == PESSIMISTIC)
                tx.nearFutureId(req.futureId());
        }
    } else {
        GridDhtPartitionTopology top = null;
        if (req.firstClientRequest()) {
            assert req.concurrency() == OPTIMISTIC : req;
            assert CU.clientNode(nearNode) : nearNode;
            top = firstEntry.context().topology();
            top.readLock();
            if (req.allowWaitTopologyFuture()) {
                GridDhtTopologyFuture topFut = top.topologyVersionFuture();
                if (!topFut.isDone()) {
                    top.readUnlock();
                    return null;
                }
            }
        }
        try {
            if (top != null) {
                boolean retry = false;
                GridDhtTopologyFuture topFut = top.topologyVersionFuture();
                if (!req.allowWaitTopologyFuture() && !topFut.isDone()) {
                    retry = true;
                    if (txPrepareMsgLog.isDebugEnabled()) {
                        txPrepareMsgLog.debug("Topology change is in progress, need remap transaction [" + "txId=" + req.version() + ", node=" + nearNode.id() + ", reqTopVer=" + req.topologyVersion() + ", locTopVer=" + top.readyTopologyVersion() + ", req=" + req + ']');
                    }
                }
                if (!retry && needRemap(req.topologyVersion(), top.readyTopologyVersion(), req)) {
                    retry = true;
                    if (txPrepareMsgLog.isDebugEnabled()) {
                        txPrepareMsgLog.debug("Topology version mismatch for near prepare, need remap transaction [" + "txId=" + req.version() + ", node=" + nearNode.id() + ", reqTopVer=" + req.topologyVersion() + ", locTopVer=" + top.readyTopologyVersion() + ", req=" + req + ']');
                    }
                }
                if (retry) {
                    GridNearTxPrepareResponse res = new GridNearTxPrepareResponse(req.partition(), req.version(), req.futureId(), req.miniId(), req.version(), req.version(), null, null, top.lastTopologyChangeVersion(), req.onePhaseCommit(), req.deployInfo() != null);
                    try {
                        ctx.io().send(nearNode, res, req.policy());
                        if (txPrepareMsgLog.isDebugEnabled()) {
                            txPrepareMsgLog.debug("Sent remap response for near prepare [txId=" + req.version() + ", node=" + nearNode.id() + ']');
                        }
                    } catch (ClusterTopologyCheckedException ignored) {
                        if (txPrepareMsgLog.isDebugEnabled()) {
                            txPrepareMsgLog.debug("Failed to send remap response for near prepare, node failed [" + "txId=" + req.version() + ", node=" + nearNode.id() + ']');
                        }
                    } catch (IgniteCheckedException e) {
                        U.error(txPrepareMsgLog, "Failed to send remap response for near prepare " + "[txId=" + req.version() + ", node=" + nearNode.id() + ", req=" + req + ']', e);
                    }
                    return new GridFinishedFuture<>(res);
                }
                assert topFut.isDone();
            }
            tx = new GridDhtTxLocal(ctx, req.topologyVersion(), nearNode.id(), req.version(), req.futureId(), req.miniId(), req.threadId(), req.implicitSingle(), req.implicitSingle(), req.system(), req.explicitLock(), req.policy(), req.concurrency(), req.isolation(), req.timeout(), req.isInvalidate(), true, req.onePhaseCommit(), req.txSize(), req.transactionNodes(), req.subjectId(), req.taskNameHash());
            tx = ctx.tm().onCreated(null, tx);
            if (tx != null)
                tx.topologyVersion(req.topologyVersion());
            else
                U.warn(log, "Failed to create local transaction (was transaction rolled back?) [xid=" + req.version() + ", req=" + req + ']');
        } finally {
            if (tx != null)
                req.txState(tx.txState());
            if (top != null)
                top.readUnlock();
        }
    }
    if (tx != null) {
        req.txState(tx.txState());
        if (req.explicitLock())
            tx.explicitLock(true);
        tx.transactionNodes(req.transactionNodes());
        if (req.near())
            tx.nearOnOriginatingNode(true);
        if (req.onePhaseCommit()) {
            assert req.last() : req;
            tx.onePhaseCommit(true);
        }
        if (req.needReturnValue())
            tx.needReturnValue(true);
        IgniteInternalFuture<GridNearTxPrepareResponse> fut = tx.prepareAsync(req);
        if (tx.isRollbackOnly() && !tx.commitOnPrepare()) {
            if (tx.state() != TransactionState.ROLLED_BACK && tx.state() != TransactionState.ROLLING_BACK)
                tx.rollbackDhtLocalAsync();
        }
        final GridDhtTxLocal tx0 = tx;
        fut.listen(new CI1<IgniteInternalFuture<?>>() {

            @Override
            public void apply(IgniteInternalFuture<?> txFut) {
                try {
                    txFut.get();
                } catch (IgniteCheckedException e) {
                    // Just in case.
                    tx0.setRollbackOnly();
                    if (!X.hasCause(e, IgniteTxOptimisticCheckedException.class) && !X.hasCause(e, IgniteFutureCancelledException.class) && !ctx.kernalContext().isStopping())
                        U.error(log, "Failed to prepare DHT transaction: " + tx0, e);
                }
            }
        });
        return fut;
    } else
        return new GridFinishedFuture<>((GridNearTxPrepareResponse) null);
}

Example 3 with GridDhtPartitionTopology

use of org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology in project ignite by apache.

the class GridDhtPartitionDemander method handleSupplyMessage.

/**
 * Handles supply message from {@code nodeId} with specified {@code topicId}.
 *
 * Supply message contains entries to populate rebalancing partitions.
 *
 * There is a cyclic process:
 * Populate rebalancing partitions with entries from Supply message.
 * If not all partitions specified in {@link #rebalanceFut} were rebalanced or marked as missed
 * send new Demand message to request next batch of entries.
 *
 * @param topicId Topic id.
 * @param nodeId Node id.
 * @param supply Supply message.
 */
public void handleSupplyMessage(int topicId, final UUID nodeId, final GridDhtPartitionSupplyMessage supply) {
    AffinityTopologyVersion topVer = supply.topologyVersion();
    final RebalanceFuture fut = rebalanceFut;
    ClusterNode node = ctx.node(nodeId);
    if (node == null)
        return;
    if (// Topology already changed (for the future that supply message based on).
    topologyChanged(fut))
        return;
    if (!fut.isActual(supply.rebalanceId())) {
        // Supple message based on another future.
        return;
    }
    if (log.isDebugEnabled())
        log.debug("Received supply message [grp=" + grp.cacheOrGroupName() + ", msg=" + supply + ']');
    // Check whether there were class loading errors on unmarshal
    if (supply.classError() != null) {
        U.warn(log, "Rebalancing from node cancelled [grp=" + grp.cacheOrGroupName() + ", node=" + nodeId + "]. Class got undeployed during preloading: " + supply.classError());
        fut.cancel(nodeId);
        return;
    }
    final GridDhtPartitionTopology top = grp.topology();
    if (grp.sharedGroup()) {
        for (GridCacheContext cctx : grp.caches()) {
            if (cctx.statisticsEnabled()) {
                long keysCnt = supply.keysForCache(cctx.cacheId());
                if (keysCnt != -1)
                    cctx.cache().metrics0().onRebalancingKeysCountEstimateReceived(keysCnt);
                // Can not be calculated per cache.
                cctx.cache().metrics0().onRebalanceBatchReceived(supply.messageSize());
            }
        }
    } else {
        GridCacheContext cctx = grp.singleCacheContext();
        if (cctx.statisticsEnabled()) {
            if (supply.estimatedKeysCount() != -1)
                cctx.cache().metrics0().onRebalancingKeysCountEstimateReceived(supply.estimatedKeysCount());
            cctx.cache().metrics0().onRebalanceBatchReceived(supply.messageSize());
        }
    }
    try {
        AffinityAssignment aff = grp.affinity().cachedAffinity(topVer);
        GridCacheContext cctx = grp.sharedGroup() ? null : grp.singleCacheContext();
        // Preload.
        for (Map.Entry<Integer, CacheEntryInfoCollection> e : supply.infos().entrySet()) {
            int p = e.getKey();
            if (aff.get(p).contains(ctx.localNode())) {
                GridDhtLocalPartition part = top.localPartition(p, topVer, true);
                assert part != null;
                boolean last = supply.last().containsKey(p);
                if (part.state() == MOVING) {
                    boolean reserved = part.reserve();
                    assert reserved : "Failed to reserve partition [igniteInstanceName=" + ctx.igniteInstanceName() + ", grp=" + grp.cacheOrGroupName() + ", part=" + part + ']';
                    part.lock();
                    try {
                        // Loop through all received entries and try to preload them.
                        for (GridCacheEntryInfo entry : e.getValue().infos()) {
                            if (!preloadEntry(node, p, entry, topVer)) {
                                if (log.isDebugEnabled())
                                    log.debug("Got entries for invalid partition during " + "preloading (will skip) [p=" + p + ", entry=" + entry + ']');
                                break;
                            }
                            if (grp.sharedGroup() && (cctx == null || cctx.cacheId() != entry.cacheId()))
                                cctx = ctx.cacheContext(entry.cacheId());
                            if (cctx != null && cctx.statisticsEnabled())
                                cctx.cache().metrics0().onRebalanceKeyReceived();
                        }
                        // then we take ownership.
                        if (last) {
                            top.own(part);
                            fut.partitionDone(nodeId, p);
                            if (log.isDebugEnabled())
                                log.debug("Finished rebalancing partition: " + part);
                        }
                    } finally {
                        part.unlock();
                        part.release();
                    }
                } else {
                    if (last)
                        fut.partitionDone(nodeId, p);
                    if (log.isDebugEnabled())
                        log.debug("Skipping rebalancing partition (state is not MOVING): " + part);
                }
            } else {
                fut.partitionDone(nodeId, p);
                if (log.isDebugEnabled())
                    log.debug("Skipping rebalancing partition (it does not belong on current node): " + p);
            }
        }
        // Only request partitions based on latest topology version.
        for (Integer miss : supply.missed()) {
            if (aff.get(miss).contains(ctx.localNode()))
                fut.partitionMissed(nodeId, miss);
        }
        for (Integer miss : supply.missed()) fut.partitionDone(nodeId, miss);
        GridDhtPartitionDemandMessage d = new GridDhtPartitionDemandMessage(supply.rebalanceId(), supply.topologyVersion(), grp.groupId());
        d.timeout(grp.config().getRebalanceTimeout());
        d.topic(rebalanceTopics.get(topicId));
        if (!topologyChanged(fut) && !fut.isDone()) {
            // Send demand message.
            try {
                ctx.io().sendOrderedMessage(node, rebalanceTopics.get(topicId), d.convertIfNeeded(node.version()), grp.ioPolicy(), grp.config().getRebalanceTimeout());
            } catch (ClusterTopologyCheckedException e) {
                if (log.isDebugEnabled()) {
                    log.debug("Node left during rebalancing [grp=" + grp.cacheOrGroupName() + ", node=" + node.id() + ", msg=" + e.getMessage() + ']');
                }
            }
        }
    } catch (IgniteSpiException | IgniteCheckedException e) {
        LT.error(log, e, "Error during rebalancing [grp=" + grp.cacheOrGroupName() + ", srcNode=" + node.id() + ", err=" + e + ']');
    }
}

Example 4 with GridDhtPartitionTopology

use of org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology in project ignite by apache.

the class GridDhtPartitionsExchangeFuture method assignPartitionsStates.

/**
 */
private void assignPartitionsStates() {
    for (Map.Entry<Integer, CacheGroupDescriptor> e : cctx.affinity().cacheGroups().entrySet()) {
        CacheGroupDescriptor grpDesc = e.getValue();
        if (grpDesc.config().getCacheMode() == CacheMode.LOCAL)
            continue;
        if (!CU.isPersistentCache(grpDesc.config(), cctx.gridConfig().getDataStorageConfiguration()))
            continue;
        CacheGroupContext grpCtx = cctx.cache().cacheGroup(e.getKey());
        GridDhtPartitionTopology top = grpCtx != null ? grpCtx.topology() : cctx.exchange().clientTopology(e.getKey(), events().discoveryCache());
        assignPartitionStates(top);
    }
}

Example 5 with GridDhtPartitionTopology

use of org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology in project ignite by apache.

the class GridCommonAbstractTest method awaitPartitionMapExchange.

/**
 * @param waitEvicts If {@code true} will wait for evictions finished.
 * @param waitNode2PartUpdate If {@code true} will wait for nodes node2part info update finished.
 * @param nodes Optional nodes. If {@code null} method will wait for all nodes, for non null collection nodes will
 *      be filtered
 * @param printPartState If {@code true} will print partition state if evictions not happened.
 * @throws InterruptedException If interrupted.
 */
@SuppressWarnings("BusyWait")
protected void awaitPartitionMapExchange(boolean waitEvicts, boolean waitNode2PartUpdate, @Nullable Collection<ClusterNode> nodes, boolean printPartState) throws InterruptedException {
    long timeout = getPartitionMapExchangeTimeout();
    long startTime = -1;
    Set<String> names = new HashSet<>();
    Ignite crd = null;
    for (Ignite g : G.allGrids()) {
        ClusterNode node = g.cluster().localNode();
        if (crd == null || node.order() < crd.cluster().localNode().order()) {
            crd = g;
            if (node.order() == 1)
                break;
        }
    }
    if (crd == null)
        return;
    AffinityTopologyVersion waitTopVer = ((IgniteKernal) crd).context().discovery().topologyVersionEx();
    if (waitTopVer.topologyVersion() <= 0)
        waitTopVer = new AffinityTopologyVersion(1, 0);
    for (Ignite g : G.allGrids()) {
        if (nodes != null && !nodes.contains(g.cluster().localNode()))
            continue;
        IgniteKernal g0 = (IgniteKernal) g;
        names.add(g0.configuration().getIgniteInstanceName());
        if (startTime != -1) {
            if (startTime != g0.context().discovery().gridStartTime())
                fail("Found nodes from different clusters, probable some test does not stop nodes " + "[allNodes=" + names + ']');
        } else
            startTime = g0.context().discovery().gridStartTime();
        if (g.cluster().localNode().isDaemon())
            continue;
        IgniteInternalFuture<?> exchFut = g0.context().cache().context().exchange().affinityReadyFuture(waitTopVer);
        if (exchFut != null && !exchFut.isDone()) {
            try {
                exchFut.get(timeout);
            } catch (IgniteCheckedException e) {
                log.error("Failed to wait for exchange [topVer=" + waitTopVer + ", node=" + g0.name() + ']', e);
            }
        }
        for (IgniteCacheProxy<?, ?> c : g0.context().cache().jcaches()) {
            CacheConfiguration cfg = c.context().config();
            if (cfg == null)
                continue;
            if (cfg.getCacheMode() != LOCAL && cfg.getRebalanceMode() != NONE && g.cluster().nodes().size() > 1) {
                AffinityFunction aff = cfg.getAffinity();
                GridDhtCacheAdapter<?, ?> dht = dht(c);
                GridDhtPartitionTopology top = dht.topology();
                for (int p = 0; p < aff.partitions(); p++) {
                    long start = 0;
                    for (int i = 0; ; i++) {
                        boolean match = false;
                        GridCachePartitionExchangeManager<?, ?> exchMgr = dht.context().shared().exchange();
                        AffinityTopologyVersion readyVer = exchMgr.readyAffinityVersion();
                        // Otherwise, there may be an assertion when printing top.readyTopologyVersion().
                        try {
                            IgniteInternalFuture<?> fut = exchMgr.affinityReadyFuture(readyVer);
                            if (fut != null)
                                fut.get();
                        } catch (IgniteCheckedException e) {
                            throw new IgniteException(e);
                        }
                        if (readyVer.topologyVersion() > 0 && c.context().started()) {
                            // Must map on updated version of topology.
                            Collection<ClusterNode> affNodes = dht.context().affinity().assignment(readyVer).idealAssignment().get(p);
                            int affNodesCnt = affNodes.size();
                            GridDhtTopologyFuture topFut = top.topologyVersionFuture();
                            Collection<ClusterNode> owners = (topFut != null && topFut.isDone()) ? top.owners(p, AffinityTopologyVersion.NONE) : Collections.<ClusterNode>emptyList();
                            int ownerNodesCnt = owners.size();
                            GridDhtLocalPartition loc = top.localPartition(p, readyVer, false);
                            if (affNodesCnt != ownerNodesCnt || !affNodes.containsAll(owners) || (waitEvicts && loc != null && loc.state() != GridDhtPartitionState.OWNING)) {
                                LT.warn(log(), "Waiting for topology map update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", p=" + p + ", affNodesCnt=" + affNodesCnt + ", ownersCnt=" + ownerNodesCnt + ", affNodes=" + F.nodeIds(affNodes) + ", owners=" + F.nodeIds(owners) + ", topFut=" + topFut + ", locNode=" + g.cluster().localNode() + ']');
                            } else
                                match = true;
                        } else {
                            LT.warn(log(), "Waiting for topology map update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", started=" + dht.context().started() + ", p=" + p + ", readVer=" + readyVer + ", locNode=" + g.cluster().localNode() + ']');
                        }
                        if (!match) {
                            if (i == 0)
                                start = System.currentTimeMillis();
                            if (System.currentTimeMillis() - start > timeout) {
                                U.dumpThreads(log);
                                if (printPartState)
                                    printPartitionState(c);
                                throw new IgniteException("Timeout of waiting for topology map update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", p=" + p + ", readVer=" + readyVer + ", locNode=" + g.cluster().localNode() + ']');
                            }
                            // Busy wait.
                            Thread.sleep(20);
                            continue;
                        }
                        if (i > 0)
                            log().warning("Finished waiting for topology map update [igniteInstanceName=" + g.name() + ", p=" + p + ", duration=" + (System.currentTimeMillis() - start) + "ms]");
                        break;
                    }
                }
                if (waitNode2PartUpdate) {
                    long start = System.currentTimeMillis();
                    boolean failed = true;
                    while (failed) {
                        failed = false;
                        for (GridDhtPartitionMap pMap : top.partitionMap(true).values()) {
                            if (failed)
                                break;
                            for (Map.Entry entry : pMap.entrySet()) {
                                if (System.currentTimeMillis() - start > timeout) {
                                    U.dumpThreads(log);
                                    throw new IgniteException("Timeout of waiting for partition state update [" + "igniteInstanceName=" + g.name() + ", cache=" + cfg.getName() + ", cacheId=" + dht.context().cacheId() + ", topVer=" + top.readyTopologyVersion() + ", locNode=" + g.cluster().localNode() + ']');
                                }
                                if (entry.getValue() != GridDhtPartitionState.OWNING) {
                                    LT.warn(log(), "Waiting for correct partition state part=" + entry.getKey() + ", should be OWNING [state=" + entry.getValue() + "], node=" + g.name() + ", cache=" + c.getName());
                                    // Busy wait.
                                    Thread.sleep(200);
                                    failed = true;
                                    break;
                                }
                            }
                        }
                    }
                }
            }
        }
    }
    log.info("awaitPartitionMapExchange finished");
}