-
Bug
-
Resolution: Done
-
Medium
-
None
-
None
-
None
while working on CONTROLLER-1831 and reviewing blueprint wiring XML in controller, I realized that opendaylight/md-sal/mdsal-trace/binding-impl/src/main/resources/org/opendaylight/blueprint/impl-blueprint.xml contained a subtle mistake:
The tracing pingpong DataBroker was, by mistake, wired to the original non-tracing
DOMDataBroker instead of the TracingDOMDataBroker. The tracing non pingpong DataBroker was already correct. This means that trace:transaction missed any leaks caused by non-closed transactions from users of the pingpong DataBroker.
Having some application code use the tracing and other the non-tracing original DB, including separate classloading due to the BindingNormalizedNodeCodecRegistry in the BindingToNormalizedNodeCodec, when the Transaction Trace tool is installed (only), smells like asking for trouble... in fact, I'm wondering if perhaps this could be causing MDSAL-213 !?
I'll raise a Gerrit with a proposed fix for this ASAP.
See also .CONTROLLER-1832
- blocks
-
NETVIRT-1089 Add trace:transactions to suite teardowns
- In Progress
-
MDSAL-213 Serializing DataObject to JSON causes frozen class exception
- Resolved
-
CONTROLLER-1831 Utilities to bootstrap CDS in a standalone environment (non-OSGi/Karaf)
- Resolved
- relates to
-
CONTROLLER-1832 Transaction Trace tool wiring creates second BindingToNormalizedNodeCodec
- Resolved