Hi, I've been looking into BZ 63931. The error messages I was expecting to see were not there.
The first odd thing I noticed was an NPE here: https://github.com/apache/tomcat/blob/master/java/org/apache/tomcat/websocket/WsRemoteEndpointImplBase.java#L310 bsh.getSendResult() returned null. That shouldn't happen. I added some debug logging and found that the call to writeMessagePart() on the previous line was returning before the SendResult was populated. Tracing the code for writeMessagePart() led me to SocketWrapperBase.vectoredOperation() The issue appears to be that state.start() at https://github.com/apache/tomcat/blob/master/java/org/apache/tomcat/util/net/SocketWrapperBase.java#L1425 effectively always does non-blocking writes. If the write buffer is full (which is the scenario I am testing) 0 bytes are written, the socket is passed to the Poller and the call returns. I hacked in the following code just after the state.start() if (block == BlockingMode.BLOCK) { if (read) { readPending.acquireUninterruptibly(); readPending.release(); } else { writePending.acquireUninterruptibly(); writePending.release(); } } I'm sure that code is wrong. It was just intended as a quick hack to force the call to block until the write had completed. It had the desired effect and I started to see the error messages I expected. I'm not familiar enough with the asyncIO code yet to know where the best place to fix this is. Rémy, are you able to help? Thanks, Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
