Pid wrote:
On 22/03/2012 15:47, Lobb, Janos wrote:
On Mar 22, 2012, at 7:40 AM, André Warnier wrote:
Martin Gainty wrote:
Mitch
One possible cause
FE Application creates session
passes queryString or posted Data (hopefully in Sesion) to backend Axis
WebService
Objects are serialized to disk with existing JSESSIONID
FE times out
back end responds to a disconnected session and tries to re-create session with
old JSESSIONID
Illegal StateException is thrown when client tries to (re)create new session
with old serialized JSESSIONID
+1
I would add that the "illegal state" indicated by the error message is a
special tomcat quantum state, resulting from a situation in which the response, while not
being totally sent yet, is in fact partially sent already, causing a certain amount of
probabilistic confusion a the level of the event horizon.
This is caught by a hidden class in Tomcat, invoking a singleton object which
writes a generally nonsensical message in one of the logs, chosen at random.
Due to the nature of the error, it is of course extremely hard to reproduce, as
one can find out either the exact time of the event, or its location, but not
both.
Even if it is an "illegal state" it still should be either "clean" or "mixed". If clean, then
Integrate(fn*(q,p)<illegal state>fn(q,p)dpdq should give the exact probability and if it is in "mixed" then
the diagonal fnm density matrix elements should do similarly. Of course if it is at the event horizon, then all bets are
off, because there is still now good theory combining quantum states with gravity :-)
All of this is irrelevant if the OP is not using the Http11QuantumConnector.
He did not give us details of his configuration, so Janos and I were talking in all
generality.
Is the Http11QuantumConnector (finally) released ? The last time I tested it, it was
starting threads in random parallel universes, which made it hard to collect the results
and clean up afterward (and never mind finding the logs). Granted, heap memory was not an
issue anymore, but still I would not call this production-level code.
I think what we really need is a network trace. But a simple back-of-the-hand calculation
shows that to determine the precise cause with a 3-sigma level of certainty, we would need
a network trace covering at least 7 X 10 exp 24 seconds, which unfortunately exceeds the
current high estimate of the age of the Universe by several magnitudes. The same
calculation shows that Konstantin's earlier explanation has a 99.87% probability of being
closer to the truth. So before setting up the network trace, we would recommend to the OP
to check Konstantin's hypothesis, just in case.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
