/**
* @param nodes Nodes.
* @param msg Message.
* @param partsMap Partitions.
* @return {@code true} If all messages sent successfully.
*/
private boolean send(
Collection<ClusterNode> nodes, Message msg, Map<ClusterNode, IntArray> partsMap) {
boolean locNodeFound = false;
boolean ok = true;
for (ClusterNode node : nodes) {
if (node.isLocal()) {
locNodeFound = true;
continue;
}
try {
ctx.io().send(node, GridTopic.TOPIC_QUERY, copy(msg, node, partsMap), QUERY_POOL);
} catch (IgniteCheckedException e) {
ok = false;
U.warn(log, e.getMessage());
}
}
if (locNodeFound) // Local node goes the last to allow parallel execution.
h2.mapQueryExecutor()
.onMessage(ctx.localNodeId(), copy(msg, ctx.discovery().localNode(), partsMap));
return ok;
}
/**
* Rolls back local part of colocated transaction.
*
* @return Commit future.
*/
public IgniteInternalFuture<IgniteInternalTx> rollbackAsyncLocal() {
if (log.isDebugEnabled()) log.debug("Rolling back colocated tx locally: " + this);
final GridDhtTxFinishFuture fut = new GridDhtTxFinishFuture<>(cctx, this, /*commit*/ false);
cctx.mvcc().addFuture(fut, fut.futureId());
IgniteInternalFuture<?> prep = prepFut.get();
if (prep == null || prep.isDone()) {
try {
if (prep != null) prep.get();
} catch (IgniteCheckedException e) {
if (log.isDebugEnabled())
log.debug(
"Failed to prepare transaction during rollback (will ignore) [tx="
+ this
+ ", msg="
+ e.getMessage()
+ ']');
}
fut.finish();
} else
prep.listen(
new CI1<IgniteInternalFuture<?>>() {
@Override
public void apply(IgniteInternalFuture<?> f) {
try {
f.get(); // Check for errors of a parent future.
} catch (IgniteCheckedException e) {
log.debug(
"Failed to prepare transaction during rollback (will ignore) [tx="
+ this
+ ", msg="
+ e.getMessage()
+ ']');
}
fut.finish();
}
});
return fut;
}
/**
* Send delete message to all meta cache nodes in the grid.
*
* @param msg Message to send.
*/
private void sendDeleteMessage(IgfsDeleteMessage msg) {
assert msg != null;
Collection<ClusterNode> nodes = meta.metaCacheNodes();
for (ClusterNode node : nodes) {
try {
igfsCtx.send(node, topic, msg, GridIoPolicy.SYSTEM_POOL);
} catch (IgniteCheckedException e) {
U.warn(
log,
"Failed to send IGFS delete message to node [nodeId="
+ node.id()
+ ", msg="
+ msg
+ ", err="
+ e.getMessage()
+ ']');
}
}
}
/**
* @param desc Process descriptor.
* @return Client.
* @throws IgniteCheckedException If failed.
*/
@Nullable
protected HadoopCommunicationClient createNioClient(HadoopProcessDescriptor desc)
throws IgniteCheckedException {
assert desc != null;
int shmemPort = desc.sharedMemoryPort();
// If remote node has shared memory server enabled and has the same set of MACs
// then we are likely to run on the same host and shared memory communication could be tried.
if (shmemPort != -1 && locProcDesc.parentNodeId().equals(desc.parentNodeId())) {
try {
return createShmemClient(desc, shmemPort);
} catch (IgniteCheckedException e) {
if (e.hasCause(IpcOutOfSystemResourcesException.class))
// Has cause or is itself the IpcOutOfSystemResourcesException.
LT.warn(log, null, OUT_OF_RESOURCES_TCP_MSG);
else if (log.isDebugEnabled())
log.debug(
"Failed to establish shared memory connection with local hadoop process: " + desc);
}
}
return createTcpClient(desc);
}
/**
* Establish TCP connection to remote hadoop process and returns client.
*
* @param desc Process descriptor.
* @return Client.
* @throws IgniteCheckedException If failed.
*/
protected HadoopCommunicationClient createTcpClient(HadoopProcessDescriptor desc)
throws IgniteCheckedException {
String addr = desc.address();
int port = desc.tcpPort();
if (log.isDebugEnabled())
log.debug(
"Trying to connect to remote process [locProcDesc="
+ locProcDesc
+ ", desc="
+ desc
+ ']');
boolean conn = false;
HadoopTcpNioCommunicationClient client = null;
IgniteCheckedException errs = null;
int connectAttempts = 1;
long connTimeout0 = connTimeout;
int attempt = 1;
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);
ch.socket().connect(new InetSocketAddress(addr, port), (int) connTimeout);
HandshakeFinish fin = new HandshakeFinish();
GridNioSession ses = nioSrvr.createSession(ch, F.asMap(HANDSHAKE_FINISH_META, fin)).get();
client = new HadoopTcpNioCommunicationClient(ses);
if (log.isDebugEnabled()) log.debug("Waiting for handshake finish for client: " + client);
fin.await(connTimeout0);
conn = true;
} catch (HadoopHandshakeTimeoutException e) {
if (client != null) {
client.forceClose();
client = null;
}
if (log.isDebugEnabled())
log.debug(
"Handshake timedout (will retry with increased timeout) [timeout="
+ connTimeout0
+ ", desc="
+ desc
+ ", port="
+ port
+ ", err="
+ e
+ ']');
if (attempt == reconCnt || connTimeout0 > maxConnTimeout) {
if (log.isDebugEnabled())
log.debug(
"Handshake timed out (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 remote Hadoop process "
+ "(is process still running?) [desc="
+ desc
+ ", addrs="
+ addr
+ ']');
errs.addSuppressed(e);
break;
} else {
attempt++;
connTimeout0 *= 2;
// Continue loop.
}
} catch (Exception e) {
if (client != null) {
client.forceClose();
client = null;
}
if (log.isDebugEnabled())
log.debug("Client creation failed [addr=" + addr + ", port=" + port + ", err=" + e + ']');
if (X.hasCause(e, SocketTimeoutException.class))
LT.warn(
log,
null,
"Connect timed out (consider increasing 'connTimeout' "
+ "configuration property) [addr="
+ addr
+ ", port="
+ port
+ ']');
if (errs == null)
errs =
new IgniteCheckedException(
"Failed to connect to remote Hadoop process (is process still running?) "
+ "[desc="
+ desc
+ ", addrs="
+ addr
+ ']');
errs.addSuppressed(e);
// Reconnect for the second time, if connection is not established.
if (connectAttempts < 2
&& (e instanceof ConnectException || X.hasCause(e, ConnectException.class))) {
connectAttempts++;
continue;
}
break;
}
}
if (client == null) {
assert errs != null;
if (X.hasCause(errs, ConnectException.class))
LT.warn(
log,
null,
"Failed to connect to a remote Hadoop process (is process still running?). "
+ "Make sure operating system firewall is disabled on local and remote host) "
+ "[addrs="
+ addr
+ ", port="
+ port
+ ']');
throw errs;
}
if (log.isDebugEnabled()) log.debug("Created client: " + client);
return client;
}
