Examples with GridNioRecoveryDescriptor org.apache.ignite.internal.util.nio.GridNioRecoveryDescriptor used on opensource projects

Example 6 with GridNioRecoveryDescriptor

use of org.apache.ignite.internal.util.nio.GridNioRecoveryDescriptor in project ignite by apache.

the class TcpCommunicationSpi method createTcpClient.

/**
     * Establish TCP connection to remote node and returns client.
     *
     * @param node Remote node.
     * @param connIdx Connection index.
     * @return Client.
     * @throws IgniteCheckedException If failed.
     */
protected GridCommunicationClient createTcpClient(ClusterNode node, int connIdx) throws IgniteCheckedException {
    Collection<String> rmtAddrs0 = node.attribute(createSpiAttributeName(ATTR_ADDRS));
    Collection<String> rmtHostNames0 = node.attribute(createSpiAttributeName(ATTR_HOST_NAMES));
    Integer boundPort = node.attribute(createSpiAttributeName(ATTR_PORT));
    Collection<InetSocketAddress> extAddrs = node.attribute(createSpiAttributeName(ATTR_EXT_ADDRS));
    boolean isRmtAddrsExist = (!F.isEmpty(rmtAddrs0) && boundPort != null);
    boolean isExtAddrsExist = !F.isEmpty(extAddrs);
    if (!isRmtAddrsExist && !isExtAddrsExist)
        throw new IgniteCheckedException("Failed to send message to the destination node. Node doesn't have any " + "TCP communication addresses or mapped external addresses. Check configuration and make sure " + "that you use the same communication SPI on all nodes. Remote node id: " + node.id());
    LinkedHashSet<InetSocketAddress> addrs;
    // Try to connect first on bound addresses.
    if (isRmtAddrsExist) {
        List<InetSocketAddress> addrs0 = new ArrayList<>(U.toSocketAddresses(rmtAddrs0, rmtHostNames0, boundPort));
        boolean sameHost = U.sameMacs(getSpiContext().localNode(), node);
        Collections.sort(addrs0, U.inetAddressesComparator(sameHost));
        addrs = new LinkedHashSet<>(addrs0);
    } else
        addrs = new LinkedHashSet<>();
    // Then on mapped external addresses.
    if (isExtAddrsExist)
        addrs.addAll(extAddrs);
    Set<InetAddress> allInetAddrs = U.newHashSet(addrs.size());
    for (InetSocketAddress addr : addrs) {
        // Skip unresolved as addr.getAddress() can return null.
        if (!addr.isUnresolved())
            allInetAddrs.add(addr.getAddress());
    }
    List<InetAddress> reachableInetAddrs = U.filterReachable(allInetAddrs);
    if (reachableInetAddrs.size() < allInetAddrs.size()) {
        LinkedHashSet<InetSocketAddress> addrs0 = U.newLinkedHashSet(addrs.size());
        List<InetSocketAddress> unreachableInetAddr = new ArrayList<>(allInetAddrs.size() - reachableInetAddrs.size());
        for (InetSocketAddress addr : addrs) {
            if (reachableInetAddrs.contains(addr.getAddress()))
                addrs0.add(addr);
            else
                unreachableInetAddr.add(addr);
        }
        addrs0.addAll(unreachableInetAddr);
        addrs = addrs0;
    }
    if (log.isDebugEnabled())
        log.debug("Addresses to connect for node [rmtNode=" + node.id() + ", addrs=" + addrs.toString() + ']');
    boolean conn = false;
    GridCommunicationClient client = null;
    IgniteCheckedException errs = null;
    int connectAttempts = 1;
    for (InetSocketAddress addr : addrs) {
        long connTimeout0 = connTimeout;
        int attempt = 1;
        IgniteSpiOperationTimeoutHelper timeoutHelper = new IgniteSpiOperationTimeoutHelper(this, !node.isClient());
        while (!conn) {
            // Reconnection on handshake timeout.
            try {
                SocketChannel ch = SocketChannel.open();
                ch.configureBlocking(true);
                ch.socket().setTcpNoDelay(tcpNoDelay);
                ch.socket().setKeepAlive(true);
                if (sockRcvBuf > 0)
                    ch.socket().setReceiveBufferSize(sockRcvBuf);
                if (sockSndBuf > 0)
                    ch.socket().setSendBufferSize(sockSndBuf);
                if (getSpiContext().node(node.id()) == null) {
                    U.closeQuiet(ch);
                    throw new ClusterTopologyCheckedException("Failed to send message " + "(node left topology): " + node);
                }
                ConnectionKey connKey = new ConnectionKey(node.id(), connIdx, -1);
                GridNioRecoveryDescriptor recoveryDesc = outRecoveryDescriptor(node, connKey);
                if (!recoveryDesc.reserve()) {
                    U.closeQuiet(ch);
                    return null;
                }
                long rcvCnt = -1;
                Map<Integer, Object> meta = new HashMap<>();
                GridSslMeta sslMeta = null;
                try {
                    ch.socket().connect(addr, (int) timeoutHelper.nextTimeoutChunk(connTimeout));
                    if (isSslEnabled()) {
                        meta.put(SSL_META.ordinal(), sslMeta = new GridSslMeta());
                        SSLEngine sslEngine = ignite.configuration().getSslContextFactory().create().createSSLEngine();
                        sslEngine.setUseClientMode(true);
                        sslMeta.sslEngine(sslEngine);
                    }
                    Integer handshakeConnIdx = connIdx;
                    rcvCnt = safeHandshake(ch, recoveryDesc, node.id(), timeoutHelper.nextTimeoutChunk(connTimeout0), sslMeta, handshakeConnIdx);
                    if (rcvCnt == -1)
                        return null;
                } finally {
                    if (recoveryDesc != null && rcvCnt == -1)
                        recoveryDesc.release();
                }
                try {
                    meta.put(CONN_IDX_META, connKey);
                    if (recoveryDesc != null) {
                        recoveryDesc.onHandshake(rcvCnt);
                        meta.put(-1, recoveryDesc);
                    }
                    GridNioSession ses = nioSrvr.createSession(ch, meta).get();
                    client = new GridTcpNioCommunicationClient(connIdx, ses, log);
                    conn = true;
                } finally {
                    if (!conn) {
                        if (recoveryDesc != null)
                            recoveryDesc.release();
                    }
                }
            } catch (HandshakeTimeoutException | IgniteSpiOperationTimeoutException e) {
                if (client != null) {
                    client.forceClose();
                    client = null;
                }
                if (failureDetectionTimeoutEnabled() && (e instanceof HandshakeTimeoutException || timeoutHelper.checkFailureTimeoutReached(e))) {
                    String msg = "Handshake timed out (failure detection timeout is reached) " + "[failureDetectionTimeout=" + failureDetectionTimeout() + ", addr=" + addr + ']';
                    onException(msg, e);
                    if (log.isDebugEnabled())
                        log.debug(msg);
                    if (errs == null)
                        errs = new IgniteCheckedException("Failed to connect to node (is node still alive?). " + "Make sure that each ComputeTask and cache Transaction has a timeout set " + "in order to prevent parties from waiting forever in case of network issues " + "[nodeId=" + node.id() + ", addrs=" + addrs + ']');
                    errs.addSuppressed(new IgniteCheckedException("Failed to connect to address: " + addr, e));
                    break;
                }
                assert !failureDetectionTimeoutEnabled();
                onException("Handshake timed out (will retry with increased timeout) [timeout=" + connTimeout0 + ", addr=" + addr + ']', e);
                if (log.isDebugEnabled())
                    log.debug("Handshake timed out (will retry with increased timeout) [timeout=" + connTimeout0 + ", addr=" + addr + ", err=" + e + ']');
                if (attempt == reconCnt || connTimeout0 > maxConnTimeout) {
                    if (log.isDebugEnabled())
                        log.debug("Handshake timedout (will stop attempts to perform the handshake) " + "[timeout=" + connTimeout0 + ", maxConnTimeout=" + maxConnTimeout + ", attempt=" + attempt + ", reconCnt=" + reconCnt + ", err=" + e.getMessage() + ", addr=" + addr + ']');
                    if (errs == null)
                        errs = new IgniteCheckedException("Failed to connect to node (is node still alive?). " + "Make sure that each ComputeTask and cache Transaction has a timeout set " + "in order to prevent parties from waiting forever in case of network issues " + "[nodeId=" + node.id() + ", addrs=" + addrs + ']');
                    errs.addSuppressed(new IgniteCheckedException("Failed to connect to address: " + addr, e));
                    break;
                } else {
                    attempt++;
                    connTimeout0 *= 2;
                // Continue loop.
                }
            } catch (Exception e) {
                if (client != null) {
                    client.forceClose();
                    client = null;
                }
                onException("Client creation failed [addr=" + addr + ", err=" + e + ']', e);
                if (log.isDebugEnabled())
                    log.debug("Client creation failed [addr=" + addr + ", err=" + e + ']');
                boolean failureDetThrReached = timeoutHelper.checkFailureTimeoutReached(e);
                if (failureDetThrReached)
                    LT.warn(log, "Connect timed out (consider increasing 'failureDetectionTimeout' " + "configuration property) [addr=" + addr + ", failureDetectionTimeout=" + failureDetectionTimeout() + ']');
                else if (X.hasCause(e, SocketTimeoutException.class))
                    LT.warn(log, "Connect timed out (consider increasing 'connTimeout' " + "configuration property) [addr=" + addr + ", connTimeout=" + connTimeout + ']');
                if (errs == null)
                    errs = new IgniteCheckedException("Failed to connect to node (is node still alive?). " + "Make sure that each ComputeTask and cache Transaction has a timeout set " + "in order to prevent parties from waiting forever in case of network issues " + "[nodeId=" + node.id() + ", addrs=" + addrs + ']');
                errs.addSuppressed(new IgniteCheckedException("Failed to connect to address: " + addr, e));
                // Reconnect for the second time, if connection is not established.
                if (!failureDetThrReached && connectAttempts < 2 && (e instanceof ConnectException || X.hasCause(e, ConnectException.class))) {
                    connectAttempts++;
                    continue;
                }
                break;
            }
        }
        if (conn)
            break;
    }
    if (client == null) {
        assert errs != null;
        if (X.hasCause(errs, ConnectException.class))
            LT.warn(log, "Failed to connect to a remote node " + "(make sure that destination node is alive and " + "operating system firewall is disabled on local and remote hosts) " + "[addrs=" + addrs + ']');
        if (getSpiContext().node(node.id()) != null && (CU.clientNode(node) || !CU.clientNode(getLocalNode())) && X.hasCause(errs, ConnectException.class, SocketTimeoutException.class, HandshakeTimeoutException.class, IgniteSpiOperationTimeoutException.class)) {
            LT.warn(log, "TcpCommunicationSpi failed to establish connection to node, node will be dropped from " + "cluster [" + "rmtNode=" + node + ", err=" + errs + ", connectErrs=" + Arrays.toString(errs.getSuppressed()) + ']');
            getSpiContext().failNode(node.id(), "TcpCommunicationSpi failed to establish connection to node [" + "rmtNode=" + node + ", errs=" + errs + ", connectErrs=" + Arrays.toString(errs.getSuppressed()) + ']');
        }
        throw errs;
    }
    return client;
}