Skip to Main Content

Java APIs

Announcement

For appeals, questions and feedback about Oracle Forums, please email oracle-forums-moderators_us@oracle.com. Technical questions should be asked in the appropriate category. Thank you!

non-blocking SocketChannel and close - huh?

843790Aug 6 2006 — edited Aug 7 2006
It looks like closing a socketchannel that is in non-blocking mode can result in a dead drop of the connection, with some bytes that have already been sent and accepted (as in, 'consumed' from the buffer) being completely dropped.

In fact, I'm generally confused: Actual C non-blocking code has a similar setup for 'close' as you can see in SocketChannel's OP_CONNECT behaviour - just because you want to connect the socket doesn't mean it magically happends without blocking. Wait for a ready flag, then try again.

It should work the same way with close (you try to close, but it may not be possible without blocking).

Is this a huge gaping bug that no one quite figured out just yet, or did I miss something? I loathe to turn on linger, as that's just a crapshoot (you never know if it actually gets through. You could run into funny surprises once the server gets a bit busy. I'd rather not) and tends to cause massive leaks on linux, at least according to some google searches.

There seems to be slightly better performance (in that I have always received all data sofar) if I close the socket instead of the channel (socketChannel.socket().close() instead of socketChannel.close()) - but this has been a random attempt at 'making it work', and I can't find any documentation that backs up that this will DEFINITELY not lose any information without letting me know somehow. That still sounds impossible with this approach.
Comments
Locked Post
New comments cannot be posted to this locked post.
Post Details
Locked on Sep 4 2006
Added on Aug 6 2006
3 comments
494 views