Hi Dejan,

This is a follow on of our conversations drop in performance when using prefetch of 0. I have not been able to reproduce that, however, I am seeing some problems when transactions are used and a large number of threads, i.e:

Test Case 1:

1. Load 100,000 3KB messages onto database - I've seen it with smaller amounts
2. Restart broker
3. Start a consumer with:

url="failover:(tcp://localhost:61637)?randomize=true&timeout=30000&jms.redeliveryPolicy.maximumRedeliveries=-1&jms.prefetchPolicy.all=0";

ant consumer -Dtransacted=true -Dbatch=1 -Durl=tcp://localhost:61637 -Dmax=1500 -DparallelThreads=10

The batch size doesn't seem to matter. This consumer will end up hanging for me (note, I didn't let it run for long time). The policy does not seem to matter but for the record we started out working with:

<policyEntry queue=">" producerFlowControl="false" prioritizedMessages="false" expireMessagesPeriod="0" optimizedDispatch="true" useCache="false" maxPageSize="10000"> </policyEntry>

Using -Dthreads=5 is OK .... I suspect the prefetch does not matter as I believe I let that default at one time as well.

I've enabled/disabled Cache and also let maxPageSize default. Same behavior. The broker appeared to be hung, see attached broker1.out.example1 thread dump and corresponding screen shot.

Tested on 5.9-SNAPSHOT and also 5.8 patched version with early drop of P3. Same behavior.