use of org.apache.ignite.spi.discovery.tcp.messages.TcpDiscoveryHandshakeResponse in project ignite by apache.
the class ServerImpl method sendMessageDirectly.
/**
* Establishes connection to an address, sends message and returns the response (if any).
*
* @param msg Message to send.
* @param addr Address to send message to.
* @param timeoutHelper Operation timeout helper.
* @return Response read from the recipient or {@code null} if no response is supposed.
* @throws IgniteSpiException If an error occurs.
*/
@Nullable
private Integer sendMessageDirectly(TcpDiscoveryAbstractMessage msg, InetSocketAddress addr, IgniteSpiOperationTimeoutHelper timeoutHelper) throws IgniteSpiException {
assert msg != null;
assert addr != null;
Collection<Throwable> errs = null;
long ackTimeout0 = spi.getAckTimeout();
int connectAttempts = 1;
int sslConnectAttempts = 3;
boolean joinReqSent;
UUID locNodeId = getLocalNodeId();
int reconCnt = 0;
while (true) {
// Need to set to false on each new iteration,
// since remote node may leave in the middle of the first iteration.
joinReqSent = false;
boolean openSock = false;
Socket sock = null;
try {
long tsNanos = System.nanoTime();
sock = spi.openSocket(addr, timeoutHelper);
openSock = true;
TcpDiscoveryHandshakeRequest req = new TcpDiscoveryHandshakeRequest(locNodeId);
// Handshake.
spi.writeToSocket(sock, req, timeoutHelper.nextTimeoutChunk(spi.getSocketTimeout()));
TcpDiscoveryHandshakeResponse res = spi.readMessage(sock, null, timeoutHelper.nextTimeoutChunk(ackTimeout0));
if (msg instanceof TcpDiscoveryJoinRequestMessage) {
boolean ignore = false;
// The only way to know is passing flag directly with handshake response.
if (!res.isDiscoveryDataPacketCompression())
((TcpDiscoveryJoinRequestMessage) msg).gridDiscoveryData().unzipData(log);
synchronized (mux) {
for (TcpDiscoveryNode failedNode : failedNodes.keySet()) {
if (failedNode.id().equals(res.creatorNodeId())) {
if (log.isDebugEnabled())
log.debug("Ignore response from node from failed list: " + res);
ignore = true;
break;
}
}
}
if (ignore)
break;
}
if (locNodeId.equals(res.creatorNodeId())) {
if (log.isDebugEnabled())
log.debug("Handshake response from local node: " + res);
break;
}
// Send message.
tsNanos = System.nanoTime();
spi.writeToSocket(sock, msg, timeoutHelper.nextTimeoutChunk(spi.getSocketTimeout()));
long tsNanos0 = System.nanoTime();
if (debugMode)
debugLog(msg, "Message has been sent directly to address [msg=" + msg + ", addr=" + addr + ", rmtNodeId=" + res.creatorNodeId() + ']');
if (log.isDebugEnabled())
log.debug("Message has been sent directly to address [msg=" + msg + ", addr=" + addr + ", rmtNodeId=" + res.creatorNodeId() + ']');
// Connection has been established, but
// join request may not be unmarshalled on remote host.
// E.g. due to class not found issue.
joinReqSent = msg instanceof TcpDiscoveryJoinRequestMessage;
int receipt = spi.readReceipt(sock, timeoutHelper.nextTimeoutChunk(ackTimeout0));
spi.stats.onMessageSent(msg, U.nanosToMillis(tsNanos0 - tsNanos));
return receipt;
} catch (ClassCastException e) {
// on dedicated machines.
if (log.isDebugEnabled())
U.error(log, "Class cast exception on direct send: " + addr, e);
onException("Class cast exception on direct send: " + addr, e);
if (errs == null)
errs = new ArrayList<>();
errs.add(e);
} catch (IOException | IgniteCheckedException e) {
if (log.isDebugEnabled())
log.error("Exception on direct send: " + e.getMessage(), e);
onException("Exception on direct send: " + e.getMessage(), e);
if (errs == null)
errs = new ArrayList<>();
errs.add(e);
if (X.hasCause(e, SSLException.class)) {
if (--sslConnectAttempts == 0)
throw new IgniteException("Unable to establish secure connection. " + "Was remote cluster configured with SSL? [rmtAddr=" + addr + ", errMsg=\"" + e.getMessage() + "\"]", e);
continue;
}
if (X.hasCause(e, StreamCorruptedException.class)) {
// StreamCorruptedException could be caused by remote node failover
if (connectAttempts < 2) {
connectAttempts++;
continue;
}
if (log.isDebugEnabled())
log.debug("Connect failed with StreamCorruptedException, skip address: " + addr);
break;
}
if (spi.failureDetectionTimeoutEnabled() && timeoutHelper.checkFailureTimeoutReached(e))
break;
if (!spi.failureDetectionTimeoutEnabled() && ++reconCnt == spi.getReconnectCount())
break;
if (!openSock) {
// Reconnect for the second time, if connection is not established.
if (connectAttempts < 2) {
connectAttempts++;
continue;
}
// Don't retry if we can not establish connection.
break;
}
if (!spi.failureDetectionTimeoutEnabled() && (e instanceof SocketTimeoutException || X.hasCause(e, SocketTimeoutException.class))) {
ackTimeout0 *= 2;
if (!checkAckTimeout(ackTimeout0))
break;
}
} finally {
U.closeQuiet(sock);
}
}
if (joinReqSent) {
if (log.isDebugEnabled())
log.debug("Join request has been sent, but receipt has not been read (returning RES_WAIT).");
// however, warning on timed out join will be output.
return RES_OK;
}
throw new IgniteSpiException("Failed to send message to address [addr=" + addr + ", msg=" + msg + ']', U.exceptionWithSuppressed("Failed to send message to address " + "[addr=" + addr + ", msg=" + msg + ']', errs));
}
use of org.apache.ignite.spi.discovery.tcp.messages.TcpDiscoveryHandshakeResponse in project ignite by apache.
the class ClientImpl method sendJoinRequest.
/**
* @param recon {@code True} if reconnects.
* @param addr Address.
* @return Socket, connect response and client acknowledge support flag.
*/
@Nullable
private T3<SocketStream, Integer, Boolean> sendJoinRequest(boolean recon, InetSocketAddress addr) {
assert addr != null;
if (log.isDebugEnabled())
log.debug("Send join request [addr=" + addr + ", reconnect=" + recon + ", locNodeId=" + getLocalNodeId() + ']');
Collection<Throwable> errs = null;
long ackTimeout0 = spi.getAckTimeout();
int reconCnt = 0;
int connectAttempts = 1;
int sslConnectAttempts = 3;
UUID locNodeId = getLocalNodeId();
IgniteSpiOperationTimeoutHelper timeoutHelper = new IgniteSpiOperationTimeoutHelper(spi, true);
DiscoveryDataPacket discoveryData = null;
while (true) {
boolean openSock = false;
Socket sock = null;
try {
long tsNanos = System.nanoTime();
sock = spi.openSocket(addr, timeoutHelper);
openSock = true;
TcpDiscoveryHandshakeRequest req = new TcpDiscoveryHandshakeRequest(locNodeId);
req.client(true);
spi.writeToSocket(sock, req, timeoutHelper.nextTimeoutChunk(spi.getSocketTimeout()));
TcpDiscoveryHandshakeResponse res = spi.readMessage(sock, null, ackTimeout0);
UUID rmtNodeId = res.creatorNodeId();
assert rmtNodeId != null;
assert !getLocalNodeId().equals(rmtNodeId);
locNode.clientRouterNodeId(rmtNodeId);
tsNanos = System.nanoTime();
TcpDiscoveryAbstractMessage msg;
if (!recon) {
TcpDiscoveryNode node = locNode;
if (locNode.order() > 0) {
node = locNode.clientReconnectNode(spi.locNodeAttrs);
marshalCredentials(node);
}
if (discoveryData == null) {
DiscoveryDataPacket dataPacket = new DiscoveryDataPacket(getLocalNodeId());
dataPacket.joiningNodeClient(true);
discoveryData = spi.collectExchangeData(dataPacket);
}
TcpDiscoveryJoinRequestMessage joinReqMsg = new TcpDiscoveryJoinRequestMessage(node, discoveryData);
TcpDiscoveryNode nodef = node;
joinReqMsg.spanContainer().span(tracing.create(TraceableMessagesTable.traceName(joinReqMsg.getClass())).addTag(SpanTags.tag(SpanTags.EVENT_NODE, SpanTags.ID), () -> nodef.id().toString()).addTag(SpanTags.tag(SpanTags.EVENT_NODE, SpanTags.CONSISTENT_ID), () -> nodef.consistentId().toString()).addLog(() -> "Created").end());
msg = joinReqMsg;
// The only way to know is passing flag directly with handshake response.
if (!res.isDiscoveryDataPacketCompression())
((TcpDiscoveryJoinRequestMessage) msg).gridDiscoveryData().unzipData(log);
} else
msg = new TcpDiscoveryClientReconnectMessage(getLocalNodeId(), rmtNodeId, lastMsgId);
msg.client(true);
if (msg instanceof TraceableMessage)
tracing.messages().beforeSend((TraceableMessage) msg);
spi.writeToSocket(sock, msg, timeoutHelper.nextTimeoutChunk(spi.getSocketTimeout()));
spi.stats.onMessageSent(msg, U.millisSinceNanos(tsNanos));
if (log.isDebugEnabled())
log.debug("Message has been sent to address [msg=" + msg + ", addr=" + addr + ", rmtNodeId=" + rmtNodeId + ']');
return new T3<>(new SocketStream(sock), spi.readReceipt(sock, timeoutHelper.nextTimeoutChunk(ackTimeout0)), res.clientAck());
} catch (IOException | IgniteCheckedException e) {
U.closeQuiet(sock);
if (log.isDebugEnabled())
log.error("Exception on joining: " + e.getMessage(), e);
onException("Exception on joining: " + e.getMessage(), e);
if (errs == null)
errs = new ArrayList<>();
errs.add(e);
if (X.hasCause(e, SSLException.class)) {
if (--sslConnectAttempts == 0)
throw new IgniteSpiException("Unable to establish secure connection. " + "Was remote cluster configured with SSL? [rmtAddr=" + addr + ", errMsg=\"" + e.getMessage() + "\"]", e);
continue;
}
if (X.hasCause(e, StreamCorruptedException.class)) {
// StreamCorruptedException could be caused by remote node failover
if (connectAttempts < 2) {
connectAttempts++;
continue;
}
if (log.isDebugEnabled())
log.debug("Connect failed with StreamCorruptedException, skip address: " + addr);
break;
}
if (timeoutHelper.checkFailureTimeoutReached(e))
break;
if (!spi.failureDetectionTimeoutEnabled() && ++reconCnt == spi.getReconnectCount())
break;
if (!openSock) {
// Reconnect for the second time, if connection is not established.
if (connectAttempts < 2) {
connectAttempts++;
continue;
}
// Don't retry if we can not establish connection.
break;
}
if (!spi.failureDetectionTimeoutEnabled() && (e instanceof SocketTimeoutException || X.hasCause(e, SocketTimeoutException.class))) {
ackTimeout0 *= 2;
if (!checkAckTimeout(ackTimeout0))
break;
}
}
}
if (log.isDebugEnabled())
log.debug("Failed to join to address [addr=" + addr + ", recon=" + recon + ", errs=" + errs + ']');
return null;
}
Aggregations