Hi graphql-java team,

Thanks for the fix in #4314.
We verified against the latest snapshot on Maven Central (0.0.0-2026-05-19T05-53-11-b46b764). The bytecode shows the new subscriptionEventExecutionDone(...) method correctly calls alternativeCallContextMap.remove(...) in both PerLevelDataLoaderDispatchStrategy and ExhaustedDataLoaderDispatchStrategy.

Production impact for us
A long-lived GraphQL subscription publisher hits OOM every couple days. The only mitigation currently available to us is periodic process restarts.

Why we can't pick up the snapshot
A timestamped snapshot isn't a viable dependency for our prod release pipeline.

Ask
Would either of these be possible?

1. Cut a 25.x patch release with Memory leak in graphql subscriptions #4314 backported.
2. Cut a 26.1 (or next stable) that includes the fix.

Either path would unblock us, and probably others.

Thanks for considering,
Kirtan