TCP flush responsibility [INT-3575]

[spring-operator]

Dennis Thrysøe opened INT-3575 and commented

A BufferedOutputStream is now given to Serializer implementations, probably as a fix for #7439 (also reported by me).

This causes the opposite problem: sending a message doesn't actually send the message, but only buffers it - causing delay under low message frequency conditions.

I have been unable to find if flushing the stream is the responsibility of the Serializer implementation. If this is not the case, then it looks like the BufferedOutputStream should be flushed in TcpNioConnection.send() and similar places.

---

Affects: 4.1 GA

Issue Links:

• Remove flush() from Serializers [INT-3577] #7540 Remove flush() from Serializers

• Nagle's algorithm only applies to subsequent writes [INT-3507] #7439 Nagle's algorithm only applies to subsequent writes

Referenced from: pull request #1338

[spring-operator]

Gary Russell commented

Yes; all the provided Serializer s flush() the output stream.

We should make a note in the migration guide and documentation that user serializers must now do the same.

[spring-operator]

Artem Bilan commented

Just Documentation issue that any custom org.springframework.core.serializer.Serializer must flush() the outputStream in the end.

[spring-operator]

Gary Russell commented

Please note that we decided to issue the flush() in the connection; this will be available in 4.1.1; the work-around for 4.1.0 is to invoke flush() in a custom serializer; Also see #7540.