Description of problem: When using a persistent store in cluster mode, if you say --no-data-dir, the wrong thing happens at init time, and the broker comes up in a confused state Version-Release number of selected component (if applicable): trunk How reproducible: always Steps to Reproduce: 1. start two brokers with a stores, specify --store-dir, and --no-data-dir 2. create some objects 3. bounce and restart a broker Actual results: broker recovers objects from the store, when it shouldn't, Expected results: should get statea from the peer. Additional info:
This is a logic flaw in the cluster store status. When --no-data-dir is specified, the logic which discards the store for the second and subsequent nodes fails to run. In addition, the cluster has nowhere to write its store status information which is normally written into the data directory. Considering that the store-dir used by the async store is unique to that store and is not available on stores generally, it makes sense therefore to make the use of --data-dir mandatory when the cluster is loaded together with any non-null store. This fix now will stop the broker with an error message if --no-data-dir is used in the presence of any non-null store. Fixed in r.905680 QA: This bug is easy to reproduce using the above steps. After the fix, the above should prevent the broker from starting with an error message.
This was checked onto the branch prior to the mrg 1.3.x rebase
(In reply to comment #2) > This was checked onto the branch prior to the mrg 1.3.x rebase Should be: This was checked onto the trunk prior to the mrg 1.3.x rebase.