Hi All, The JobResultStore interface includes an Abstract implementation known as `AbstractThreadsafeJobResultStore`. This implementation is extended by two concrete implementations: `FileSystemJobResultStore` and `EmbeddedJobResultStore`. Both of these concrete implementations handle two types of job results: clean and dirty. The `EmbeddedJobResultStore` is an in-memory, thread-safe implementation for managing job results. It currently allows the storage of both clean and dirty jobs for further use. However, it retains entries for clean job results indefinitely, which can lead to unnecessary memory consumption.
To address this, we propose adding a new configuration option to allow users to define a TTL (time-to-live) for map entries. We are eager to hear feedback and thoughts from the Flink community regarding introducing new public config. During the review of PR[1] corresponding to JIRA[2], Matthias Pohl had an idea to extend the clean-up functionality to both `FileSystemJobResultStore` and `EmbeddedJobResultStore`.(Please check review comments in PR[1] for more detail) If the community finds this enhancement valuable, we plan to work on implementing this clean-up improvement and potentially introduce a FLIP (Flink Improvement Proposal) if required. Your insights and opinions on this proposal are highly appreciated. Bests, Samrat [1] https://github.com/apache/flink/pull/23531 [2] https://issues.apache.org/jira/browse/FLINK-32806