[
https://issues.apache.org/jira/browse/CAMEL-24145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18097052#comment-18097052
]
Claus Ibsen commented on CAMEL-24145:
-------------------------------------
Fixed via https://github.com/apache/camel/pull/24829
> camel-saga: InMemorySagaCoordinator compensate()/complete() return
> already-completed futures - finalization failures never propagate
> ------------------------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-24145
> URL: https://issues.apache.org/jira/browse/CAMEL-24145
> Project: Camel
> Issue Type: Bug
> Components: camel-core
> Reporter: Federico Mariani
> Assignee: Claus Ibsen
> Priority: Major
> Fix For: 4.22.0
>
>
> h3. Problem
> {{InMemorySagaCoordinator.compensate()}} and {{complete()}}
> (core/camel-support/src/main/java/org/apache/camel/saga/InMemorySagaCoordinator.java:122-155)
> kick off {{doCompensate()}}/{{doComplete()}} - an async chain running on the
> saga executor - and then unconditionally:
> {code:java}
> return CompletableFuture.completedFuture(null);
> {code}
> h3. Consequence
> * The error handling written around these calls is dead code with the
> in-memory service: {{SagaProcessor.handleSagaCompletion}} does
> {{coordinator.complete(exchange).whenComplete((done, ex) ->
> ifNotException(ex, ...))}} (SagaProcessor.java:89-101), and {{SagaProducer}}
> ({{saga:complete}}/{{saga:compensate}}) does the same - but {{ex}} can never
> be non-null and the future is already done. The exchange finishes before any
> completion/compensation endpoint has been invoked even once, and a completion
> endpoint that fails all retries never fails or flags anything beyond a WARN
> log.
> * Semantics diverge from camel-lra, where {{complete()}}/{{compensate()}}
> return the coordinator HTTP call future and a rejected close/cancel fails the
> exchange.
> * Callers of a saga route get their reply before the completion callbacks ran
> - an observable ordering difference vs LRA and a source of test flakiness.
> h3. Suggested fix
> Deliberate asynchrony is defensible for the compensation *retries* (5 x 5s by
> default must not block the failing exchange), but the returned future should
> at least reflect the outcome of the first finalization attempt (or the state
> transition validation), and the async contract of {{CamelSagaCoordinator}}
> should be documented so both implementations agree.
> Found during a broader Saga EIP / camel-lra code review (July 2026).
> _Claude Code on behalf of Croway_
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
