Transaction Timeout behaviour

Question

I am doing some testing to reproduce a production scenario where we ended up with messages stuck in JMS queue in receive state ( weblogic 10.3.3)...
The flow is as below:
JMS Queue --> Transactional MDB with resource reference
The MDB makes a callout to http service . No read time out configured for the http invocation
Transaction Timeout : 10 minutes for the MDB
I put a Thread.sleep in the target http service for 11 minutes
Tested by placing a message in the queue. MDB picked the message and made http call. Message substate was receive-transaction. Queue had redelivery delay of 55 seconds and redelivery limit of -1.
I would like to know what is the designed behaviour in this scenario when transaction timeout occurs at 10 minutes. ?
The behaviour I observed,
From JTA Monitoring ,At 602nd second, transaction was marked for rolled back, though the MDB code was executing [ I think this is the correct behaviuor]
Message substate was receieve-transaction till end of 660th second, when it got redelivered. Redelivery count increased by 1. I was expecting message substate to change to delayed, but it didn't,. New substate after 660th second was also receive transaction. In JTA monitoring I could also see a new transaction spawning up.
The whole purpose of my experiment was to reproduce the scenrio we are observing messages in 'receive' state when http service goes out. Any suggestion how i can get this, all I could get is receive transaction.
Thanks In Advance.

Answer 1

So if the delay is set to 11 minutes, but the message is trapped in a receive transaction that times out after 10 minutes, there will only be a 1 minute redelivery delay... Thanks Tom. This is eaxctly the behaviour I am observing. And if redelivery delay is less than transaction timeout period, message is redelivered instantaneously when the transaction timeout occurs.
It seems you're looking into the final details of the "redelivery delay" feature's behavior? I am looking at how to trap the message in receive state. We are seeing messages stuck in receive state in production when http service goes out and it occurs randomly. My guess is since we have not configured http read time out, the socket timeout occurs at a time beyond transaction timeout and some of the complex underlying behaviours at transaction timeout event is the reason for this [ could even be a bug when dealing with this abnormal condition].
A solution for this is to set out the http read time out less than transaction timeout. We have done this in prod now and since we cant reproduce the issue , we dont know whether it fixes the problem. We are now waiting to see if the issue reoccurs . The customer doesn't like this and wants us to reproduce the issue for confirmation.
I am looking at the message substate transitions. I know for transactional weblogic MDB's and non UOO messages, the state transition will be :
Visible --> receive --> receive transaction --> transaction committed or rolled back
In this transition , it is possible for messge in receive transaction to revert back to receive status ?
P.S. we have spent huge amount of time and effort with Oracle support with no real solutions. They want us to tell them the method to reproduce the issue.