Andrew Kyle Purtell created PHOENIX-7817:
--------------------------------------------
Summary: OMID transaction singleton view to survive cluster cycles
Key: PHOENIX-7817
URL: https://issues.apache.org/jira/browse/PHOENIX-7817
Project: Phoenix
Issue Type: Sub-task
Components: omid, test
Reporter: Andrew Kyle Purtell
Assignee: Andrew Kyle Purtell
Fix For: 5.4.0, 5.3.1
{{OmidTransactionClient}} captures the provider's {{HBaseTransactionManager}}
at construction time. {{OmidTransactionContext}} in turn caches that same
reference in a private field. When a prior test class shuts down OMID's TSO
together with its mini-cluster, the next test class reuses the stale wrapper
and is dispatched to the dead executor. The symptom is a
{{RejectedExecutionException}} originating from the OMID {{TSOClient}}
state-machine executor.
In production the singleton's manager is set exactly once and never replaced,
but the test framework exercises the swap path between classes within a single
Surefire/Failsafe fork.
Fix:
{{OmidTransactionProvider.OmidTransactionClient}} becomes a thin view onto the
singleton's current {{{}HBaseTransactionManager{}}}. It resolves
{{provider.transactionManager}} on every call to
{{{}getTransactionClient(){}}}. {{transactionManager}} is marked volatile so
test-framework injection is published to other threads without synchronization.
In production the dynamic lookup is functionally identical to caching.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
