Examples with IgniteSpiException org.apache.ignite.spi.IgniteSpiException used on opensource projects

Example 16 with IgniteSpiException

use of org.apache.ignite.spi.IgniteSpiException in project ignite by apache.

the class GridDhtPartitionSupplier method handleDemandMessage.

/**
 * For each demand message method lookups (or creates new) supply context and starts to iterate entries across requested partitions.
 * Each entry in iterator is placed to prepared supply message.
 *
 * If supply message size in bytes becomes greater than {@link IgniteConfiguration#getRebalanceBatchSize()}
 * method sends this message to demand node and saves partial state of iterated entries to supply context,
 * then restores the context again after new demand message with the same context id is arrived.
 *
 * @param topicId Id of the topic is used for the supply-demand communication.
 * @param nodeId Id of the node which sent the demand message.
 * @param demandMsg Demand message.
 */
public void handleDemandMessage(int topicId, UUID nodeId, GridDhtPartitionDemandMessage demandMsg) {
    assert demandMsg != null;
    assert nodeId != null;
    T3<UUID, Integer, AffinityTopologyVersion> contextId = new T3<>(nodeId, topicId, demandMsg.topologyVersion());
    if (demandMsg.rebalanceId() < 0) {
        // Demand node requested context cleanup.
        synchronized (scMap) {
            SupplyContext sctx = scMap.get(contextId);
            if (sctx != null && sctx.rebalanceId == -demandMsg.rebalanceId()) {
                clearContext(scMap.remove(contextId), log);
                if (log.isDebugEnabled())
                    log.debug("Supply context cleaned [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ", supplyContext=" + sctx + "]");
            } else {
                if (log.isDebugEnabled())
                    log.debug("Stale supply context cleanup message [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ", supplyContext=" + sctx + "]");
            }
            return;
        }
    }
    ClusterNode demanderNode = grp.shared().discovery().node(nodeId);
    if (demanderNode == null) {
        if (log.isDebugEnabled())
            log.debug("Demand message rejected (demander left cluster) [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + "]");
        return;
    }
    IgniteRebalanceIterator iter = null;
    SupplyContext sctx = null;
    Set<Integer> remainingParts = null;
    GridDhtPartitionSupplyMessage supplyMsg = new GridDhtPartitionSupplyMessage(demandMsg.rebalanceId(), grp.groupId(), demandMsg.topologyVersion(), grp.deploymentEnabled());
    try {
        synchronized (scMap) {
            sctx = scMap.remove(contextId);
            if (sctx != null && demandMsg.rebalanceId() < sctx.rebalanceId) {
                // Stale message, return context back and return.
                scMap.put(contextId, sctx);
                if (log.isDebugEnabled())
                    log.debug("Stale demand message [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ", actualContext=" + sctx + "]");
                return;
            }
        }
        // Demand request should not contain empty partitions if no supply context is associated with it.
        if (sctx == null && (demandMsg.partitions() == null || demandMsg.partitions().isEmpty())) {
            if (log.isDebugEnabled())
                log.debug("Empty demand message (no context and partitions) [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + "]");
            return;
        }
        if (log.isDebugEnabled())
            log.debug("Demand message accepted [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + "]");
        assert !(sctx != null && !demandMsg.partitions().isEmpty());
        long maxBatchesCnt = /* Each thread should gain prefetched batches. */
        grp.preloader().batchesPrefetchCount() * grp.shared().gridConfig().getRebalanceThreadPoolSize();
        if (sctx == null) {
            if (log.isDebugEnabled())
                log.debug("Starting supplying rebalancing [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ", fullPartitions=" + S.compact(demandMsg.partitions().fullSet()) + ", histPartitions=" + S.compact(demandMsg.partitions().historicalSet()) + "]");
        } else
            maxBatchesCnt = 1;
        if (sctx == null || sctx.iterator == null) {
            remainingParts = new HashSet<>(demandMsg.partitions().fullSet());
            CachePartitionPartialCountersMap histMap = demandMsg.partitions().historicalMap();
            for (int i = 0; i < histMap.size(); i++) {
                int p = histMap.partitionAt(i);
                remainingParts.add(p);
            }
            iter = grp.offheap().rebalanceIterator(demandMsg.partitions(), demandMsg.topologyVersion());
            for (Integer part : demandMsg.partitions().fullSet()) {
                if (iter.isPartitionMissing(part))
                    continue;
                GridDhtLocalPartition loc = top.localPartition(part, demandMsg.topologyVersion(), false);
                assert loc != null && loc.state() == GridDhtPartitionState.OWNING : "Partition should be in OWNING state: " + loc;
                supplyMsg.addEstimatedKeysCount(loc.dataStore().fullSize());
            }
            for (int i = 0; i < histMap.size(); i++) {
                int p = histMap.partitionAt(i);
                if (iter.isPartitionMissing(p))
                    continue;
                supplyMsg.addEstimatedKeysCount(histMap.updateCounterAt(i) - histMap.initialUpdateCounterAt(i));
            }
        } else {
            iter = sctx.iterator;
            remainingParts = sctx.remainingParts;
        }
        final int msgMaxSize = grp.preloader().batchSize();
        long batchesCnt = 0;
        CacheDataRow prevRow = null;
        while (iter.hasNext()) {
            CacheDataRow row = iter.peek();
            // Prevent mvcc entry history splitting into separate batches.
            boolean canFlushHistory = !grp.mvccEnabled() || prevRow != null && ((grp.sharedGroup() && row.cacheId() != prevRow.cacheId()) || !row.key().equals(prevRow.key()));
            if (canFlushHistory && supplyMsg.messageSize() >= msgMaxSize) {
                if (++batchesCnt >= maxBatchesCnt) {
                    saveSupplyContext(contextId, iter, remainingParts, demandMsg.rebalanceId());
                    reply(topicId, demanderNode, demandMsg, supplyMsg, contextId);
                    return;
                } else {
                    if (!reply(topicId, demanderNode, demandMsg, supplyMsg, contextId))
                        return;
                    supplyMsg = new GridDhtPartitionSupplyMessage(demandMsg.rebalanceId(), grp.groupId(), demandMsg.topologyVersion(), grp.deploymentEnabled());
                }
            }
            row = iter.next();
            prevRow = row;
            int part = row.partition();
            GridDhtLocalPartition loc = top.localPartition(part, demandMsg.topologyVersion(), false);
            assert (loc != null && loc.state() == OWNING && loc.reservations() > 0) || iter.isPartitionMissing(part) : "Partition should be in OWNING state and has at least 1 reservation " + loc;
            if (iter.isPartitionMissing(part) && remainingParts.contains(part)) {
                supplyMsg.missed(part);
                remainingParts.remove(part);
                if (grp.eventRecordable(EVT_CACHE_REBALANCE_PART_MISSED))
                    grp.addRebalanceMissEvent(part);
                if (log.isDebugEnabled())
                    log.debug("Requested partition is marked as missing [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ", p=" + part + "]");
                continue;
            }
            if (!remainingParts.contains(part))
                continue;
            GridCacheEntryInfo info = extractEntryInfo(row);
            if (info == null)
                continue;
            supplyMsg.addEntry0(part, iter.historical(part), info, grp.shared(), grp.cacheObjectContext());
            if (iter.isPartitionDone(part)) {
                supplyMsg.last(part, loc.updateCounter());
                remainingParts.remove(part);
                if (grp.eventRecordable(EVT_CACHE_REBALANCE_PART_SUPPLIED))
                    grp.addRebalanceSupplyEvent(part);
            }
        }
        Iterator<Integer> remainingIter = remainingParts.iterator();
        while (remainingIter.hasNext()) {
            int p = remainingIter.next();
            if (iter.isPartitionDone(p)) {
                GridDhtLocalPartition loc = top.localPartition(p, demandMsg.topologyVersion(), false);
                assert loc != null : "Supply partition is gone: grp=" + grp.cacheOrGroupName() + ", p=" + p;
                supplyMsg.last(p, loc.updateCounter());
                remainingIter.remove();
                if (grp.eventRecordable(EVT_CACHE_REBALANCE_PART_SUPPLIED))
                    grp.addRebalanceSupplyEvent(p);
            } else if (iter.isPartitionMissing(p)) {
                supplyMsg.missed(p);
                remainingIter.remove();
                if (grp.eventRecordable(EVT_CACHE_REBALANCE_PART_MISSED))
                    grp.addRebalanceMissEvent(p);
            }
        }
        assert remainingParts.isEmpty() : "Partitions after rebalance should be either done or missing: " + remainingParts;
        if (sctx != null)
            clearContext(sctx, log);
        else
            iter.close();
        reply(topicId, demanderNode, demandMsg, supplyMsg, contextId);
        if (log.isInfoEnabled())
            log.info("Finished supplying rebalancing [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + "]");
    } catch (Throwable t) {
        if (iter != null && !iter.isClosed()) {
            try {
                iter.close();
            } catch (IgniteCheckedException e) {
                t.addSuppressed(e);
            }
        }
        if (grp.shared().kernalContext().isStopping())
            return;
        // Sending supply messages with error requires new protocol.
        boolean sendErrMsg = demanderNode.version().compareTo(GridDhtPartitionSupplyMessageV2.AVAILABLE_SINCE) >= 0;
        if (t instanceof IgniteSpiException) {
            if (log.isDebugEnabled())
                log.debug("Failed to send message to node (current node is stopping?) [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ", msg=" + t.getMessage() + ']');
            sendErrMsg = false;
        } else
            U.error(log, "Failed to continue supplying [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ']', t);
        try {
            if (sctx != null)
                clearContext(sctx, log);
        } catch (Throwable t1) {
            U.error(log, "Failed to cleanup supplying context [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ']', t1);
        }
        if (!sendErrMsg)
            return;
        boolean fallbackToFullRebalance = X.hasCause(t, IgniteHistoricalIteratorException.class);
        try {
            GridDhtPartitionSupplyMessage errMsg;
            if (fallbackToFullRebalance) {
                // Mark the last checkpoint as not applicable for WAL rebalance.
                grp.shared().database().lastCheckpointInapplicableForWalRebalance(grp.groupId());
                // Mark all remaining partitions as missed to trigger full rebalance.
                if (iter == null && F.isEmpty(remainingParts)) {
                    remainingParts = new HashSet<>(demandMsg.partitions().fullSet());
                    remainingParts.addAll(demandMsg.partitions().historicalSet());
                }
                for (int p : Optional.ofNullable(remainingParts).orElseGet(Collections::emptySet)) supplyMsg.missed(p);
                errMsg = supplyMsg;
            } else {
                errMsg = new GridDhtPartitionSupplyMessageV2(demandMsg.rebalanceId(), grp.groupId(), demandMsg.topologyVersion(), grp.deploymentEnabled(), t);
            }
            reply(topicId, demanderNode, demandMsg, errMsg, contextId);
        } catch (Throwable t1) {
            U.error(log, "Failed to send supply error message [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ']', t1);
        }
        // instead of triggering failure handler.
        if (!fallbackToFullRebalance) {
            grp.shared().kernalContext().failure().process(new FailureContext(FailureType.CRITICAL_ERROR, new IgniteCheckedException("Failed to continue supplying [" + supplyRoutineInfo(topicId, nodeId, demandMsg) + ']', t)));
        }
    }
}

Example 17 with IgniteSpiException

use of org.apache.ignite.spi.IgniteSpiException in project ignite by apache.

the class IndexProcessor method store.

/**
 * Updates index with new row. Note that key is unique for cache, so if cache contains multiple indexes
 * the key should be removed from indexes other than one being updated.
 *
 * @param idxs List of indexes to update.
 * @param newRow cache row to store in index.
 * @param prevRow optional cache row that will be replaced with new row.
 */
public void store(Collection<? extends Index> idxs, CacheDataRow newRow, @Nullable CacheDataRow prevRow, boolean prevRowAvailable) throws IgniteSpiException {
    IgniteCheckedException err = null;
    ddlLock.readLock().lock();
    try {
        for (Index idx : idxs) err = updateIndex(idx, newRow, prevRow, prevRowAvailable, err);
        if (err != null)
            throw err;
    } catch (IgniteCheckedException e) {
        throw new IgniteSpiException("Failed to store row in index", e);
    } finally {
        ddlLock.readLock().unlock();
    }
}

Example 18 with IgniteSpiException

use of org.apache.ignite.spi.IgniteSpiException in project ignite by apache.

the class TcpDiscoverySpi method initializeImpl.

/**
 */
private void initializeImpl() {
    if (impl != null)
        return;
    sslEnable = ignite().configuration().getSslContextFactory() != null;
    initFailureDetectionTimeout();
    if (!forceSrvMode && (Boolean.TRUE.equals(ignite.configuration().isClientMode()))) {
        if (ackTimeout == 0)
            ackTimeout = DFLT_ACK_TIMEOUT_CLIENT;
        if (sockTimeout == 0)
            sockTimeout = DFLT_SOCK_TIMEOUT_CLIENT;
        impl = new ClientImpl(this);
        ctxInitLatch.countDown();
    } else {
        if (ackTimeout == 0)
            ackTimeout = DFLT_ACK_TIMEOUT;
        if (sockTimeout == 0)
            sockTimeout = DFLT_SOCK_TIMEOUT;
        impl = new ServerImpl(this);
    }
    metricsUpdateFreq = ignite.configuration().getMetricsUpdateFrequency();
    if (!failureDetectionTimeoutEnabled()) {
        assertParameter(sockTimeout > 0, "sockTimeout > 0");
        assertParameter(ackTimeout > 0, "ackTimeout > 0");
        assertParameter(maxAckTimeout > ackTimeout, "maxAckTimeout > ackTimeout");
        assertParameter(reconCnt > 0, "reconnectCnt > 0");
    }
    assertParameter(netTimeout > 0, "networkTimeout > 0");
    assertParameter(ipFinder != null, "ipFinder != null");
    assertParameter(metricsUpdateFreq > 0, "metricsUpdateFreq > 0" + " (inited from igniteConfiguration.metricsUpdateFrequency)");
    assertParameter(ipFinderCleanFreq > 0, "ipFinderCleanFreq > 0");
    assertParameter(locPort > 1023, "localPort > 1023");
    assertParameter(locPortRange >= 0, "localPortRange >= 0");
    assertParameter(locPort + locPortRange <= 0xffff, "locPort + locPortRange <= 0xffff");
    assertParameter(threadPri > 0, "threadPri > 0");
    assertParameter(statsPrintFreq >= 0, "statsPrintFreq >= 0");
    if (isSslEnabled()) {
        try {
            SSLContext sslCtx = ignite().configuration().getSslContextFactory().create();
            sslSockFactory = sslCtx.getSocketFactory();
            sslSrvSockFactory = sslCtx.getServerSocketFactory();
        } catch (IgniteException e) {
            throw new IgniteSpiException("Failed to create SSL context. SSL factory: " + ignite.configuration().getSslContextFactory(), e);
        }
    }
    try {
        locHost = U.resolveLocalHost(locAddr);
    } catch (IOException e) {
        throw new IgniteSpiException("Unknown local address: " + locAddr, e);
    }
    if (log.isDebugEnabled()) {
        log.debug(configInfo("localHost", locHost.getHostAddress()));
        log.debug(configInfo("localPort", locPort));
        log.debug(configInfo("localPortRange", locPortRange));
        log.debug(configInfo("threadPri", threadPri));
        if (!failureDetectionTimeoutEnabled()) {
            log.debug("Failure detection timeout is ignored because at least one of the parameters from this list" + " has been set explicitly: 'sockTimeout', 'ackTimeout', 'maxAckTimeout', 'reconnectCount'.");
            log.debug(configInfo("networkTimeout", netTimeout));
            log.debug(configInfo("sockTimeout", sockTimeout));
            log.debug(configInfo("ackTimeout", ackTimeout));
            log.debug(configInfo("maxAckTimeout", maxAckTimeout));
            log.debug(configInfo("reconnectCount", reconCnt));
        } else
            log.debug(configInfo("failureDetectionTimeout", failureDetectionTimeout()));
        log.debug(configInfo("ipFinder", ipFinder));
        log.debug(configInfo("ipFinderCleanFreq", ipFinderCleanFreq));
        log.debug(configInfo("metricsUpdateFreq", metricsUpdateFreq));
        log.debug(configInfo("statsPrintFreq", statsPrintFreq));
    }
    // Warn on odd network timeout.
    if (netTimeout < 3000)
        U.warn(log, "Network timeout is too low (at least 3000 ms recommended): " + netTimeout);
    if (ipFinder instanceof TcpDiscoveryMulticastIpFinder) {
        TcpDiscoveryMulticastIpFinder mcastIpFinder = ((TcpDiscoveryMulticastIpFinder) ipFinder);
        if (mcastIpFinder.getLocalAddress() == null)
            mcastIpFinder.setLocalAddress(locAddr);
    }
    cfgNodeId = ignite.configuration().getNodeId();
}

Example 19 with IgniteSpiException

use of org.apache.ignite.spi.IgniteSpiException in project ignite by apache.

the class ServerImpl method sendCustomEvent.

/**
 * {@inheritDoc}
 */
@Override
public void sendCustomEvent(DiscoverySpiCustomMessage evt) {
    try {
        TcpDiscoveryCustomEventMessage msg;
        if (((CustomMessageWrapper) evt).delegate() instanceof DiscoveryServerOnlyCustomMessage)
            msg = new TcpDiscoveryServerOnlyCustomEventMessage(getLocalNodeId(), evt, U.marshal(spi.marshaller(), evt));
        else
            msg = new TcpDiscoveryCustomEventMessage(getLocalNodeId(), evt, U.marshal(spi.marshaller(), evt));
        Span rootSpan = tracing.create(TraceableMessagesTable.traceName(msg.getClass())).addTag(SpanTags.tag(SpanTags.EVENT_NODE, SpanTags.ID), () -> getLocalNodeId().toString()).addTag(SpanTags.tag(SpanTags.EVENT_NODE, SpanTags.CONSISTENT_ID), () -> locNode.consistentId().toString()).addTag(SpanTags.MESSAGE_CLASS, () -> ((CustomMessageWrapper) evt).delegate().getClass().getSimpleName()).addLog(() -> "Created");
        // This root span will be parent both from local and remote nodes.
        msg.spanContainer().serializedSpanBytes(tracing.serialize(rootSpan));
        msgWorker.addMessage(msg);
        rootSpan.addLog(() -> "Sent").end();
    } catch (IgniteCheckedException e) {
        throw new IgniteSpiException("Failed to marshal custom event: " + evt, e);
    }
}

Example 20 with IgniteSpiException

use of org.apache.ignite.spi.IgniteSpiException in project ignite by apache.

the class ClientImpl method sendCustomEvent.

/**
 * {@inheritDoc}
 */
@Override
public void sendCustomEvent(DiscoverySpiCustomMessage evt) {
    State state = this.state;
    if (state == DISCONNECTED)
        throw new IgniteClientDisconnectedException(null, "Failed to send custom message: client is disconnected.");
    if (state == STOPPED || state == SEGMENTED || state == STARTING)
        throw new IgniteException("Failed to send custom message: client is " + state.name().toLowerCase() + ".");
    try {
        TcpDiscoveryCustomEventMessage msg;
        if (((CustomMessageWrapper) evt).delegate() instanceof DiscoveryServerOnlyCustomMessage)
            msg = new TcpDiscoveryServerOnlyCustomEventMessage(getLocalNodeId(), evt, U.marshal(spi.marshaller(), evt));
        else
            msg = new TcpDiscoveryCustomEventMessage(getLocalNodeId(), evt, U.marshal(spi.marshaller(), evt));
        Span rootSpan = tracing.create(TraceableMessagesTable.traceName(msg.getClass())).addTag(SpanTags.tag(SpanTags.EVENT_NODE, SpanTags.ID), () -> getLocalNodeId().toString()).addTag(SpanTags.tag(SpanTags.EVENT_NODE, SpanTags.CONSISTENT_ID), () -> locNode.consistentId().toString()).addTag(SpanTags.MESSAGE_CLASS, () -> ((CustomMessageWrapper) evt).delegate().getClass().getSimpleName()).addLog(() -> "Created");
        // This root span will be parent both from local and remote nodes.
        msg.spanContainer().serializedSpanBytes(tracing.serialize(rootSpan));
        sockWriter.sendMessage(msg);
        rootSpan.addLog(() -> "Sent").end();
    } catch (IgniteCheckedException e) {
        throw new IgniteSpiException("Failed to marshal custom event: " + evt, e);
    }
}