Github user zentol commented on a diff in the pull request:
https://github.com/apache/flink/pull/5834#discussion_r184030230
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/taskexecutor/TaskManagerRunner.java
---
@@ -355,13 +359,53 @@ public static RpcService createRpcService(
taskManagerHostname,
taskManagerAddress.getHostAddress());
}
- final int rpcPort =
configuration.getInteger(ConfigConstants.TASK_MANAGER_IPC_PORT_KEY, 0);
+ final String portRangeDefinition =
configuration.getString(TaskManagerOptions.RPC_PORT, "0");
- checkState(rpcPort >= 0 && rpcPort <= 65535, "Invalid value for
" +
- "'%s' (port for the TaskManager actor system) :
%d - Leave config parameter empty or " +
- "use 0 to let the system choose port
automatically.",
- ConfigConstants.TASK_MANAGER_IPC_PORT_KEY, rpcPort);
+ // parse port range definition and create port iterator
+ Iterator<Integer> portsIterator;
+ try {
+ portsIterator =
NetUtils.getPortRangeFromString(portRangeDefinition);
+ } catch (Exception e) {
+ throw new IllegalArgumentException("Invalid port range
definition: " + portRangeDefinition);
+ }
+
+ while (portsIterator.hasNext()) {
+ // first, we check if the port is available by opening
a socket
+ // if the actor system fails to start on the port, we
try further
+ ServerSocket availableSocket =
NetUtils.createSocketFromPorts(
+ portsIterator,
+ new NetUtils.SocketFactory() {
+ @Override
+ public ServerSocket createSocket(int
port) throws IOException {
+ return new ServerSocket(port);
+ }
+ });
+
+ int port;
+ if (availableSocket == null) {
+ throw new BindException("Unable to allocate
further port in port range: " + portRangeDefinition);
+ } else {
+ port = availableSocket.getLocalPort();
+ try {
+ availableSocket.close();
+ } catch (IOException ignored) {}
+ }
+
+ try {
+ return
AkkaRpcServiceUtils.createRpcService(taskManagerHostname, port, configuration);
+ }
+ catch (Exception e) {
+ // we can continue to try if this contains a
netty channel exception
+ Throwable cause = e.getCause();
+ if (!(cause instanceof
org.jboss.netty.channel.ChannelException ||
+ cause instanceof
java.net.BindException)) {
+ throw e;
+ } // else fall through the loop and try the
next port
+ }
+ }
- return
AkkaRpcServiceUtils.createRpcService(taskManagerHostname, rpcPort,
configuration);
+ // if we come here, we have exhausted the port range
+ throw new BindException("Could not start actor system on any
port in port range "
--- End diff --
should not mention actor system but taskmanager instead
---
