I don't have any answers, just some observation and speculation.
On 18.04.2008, at 10:37, Daniel Janus wrote:
[...]
I did some preliminary investigation and found a workaround to the problem. It seems that building up the whole XML-RPC request beforehand, then sending it all at once (rather than FORMATting the HTTP headers to the TCP stream one-by-one) fixes the problem. This is what the attached patch does. I enclose it just in case someone else runs into the same problem; it works for me, on the downside, it probably causes the library to do more consing, although I did not check that empirically.
One difference between these two cases could be that the request is sent in multiple network packets in the first case, as a single packet in the second case (if it's not too big).
Strangely enough, the problem does not seem to occur with an installation of the very same Java application on a single (non-clustered) Glassfish running on newer versions of Ubuntu.
Here I speculate wildly and without any evidence that the request consisting of multiple packets gets load-balanced across different server instances; I don't know how to verify that, though.
Cheers,
Rudi