DanielLeens commented on issue #12456: URL: https://github.com/apache/seatunnel/issues/12456#issuecomment-5869563283
Classification: A / Zeta class-loader lifecycle and Metaspace. The new before-and-after snapshots materially improve the evidence: loaders appear at job submission and stop growing while idle, and the finished MongoDB-to-Hive job plus MAT paths point to Hadoop static registries retaining classes loaded by SeaTunnelChildFirstClassLoader. They do not yet prove whether the engine failed to reach a zero reference release or whether a post-release static or thread reference is the retaining root. Current DefaultClassLoaderService already removes a zero-reference non-cache entry and clears matching thread context class loaders, so a per-registry cleanup patch or a broad parent-first Hadoop or Parquet change would be premature and could create dependency-isolation regressions. Please next provide one sanitized, time-aligned run on the affected worker: Create and Release classloader log lines for each JAR key, the reference-count result after the finished job, and MAT path-to-GC-roots for one loader after that release. Pair this with a minimal repeated MongoDB-to-Hive reproduction under cache mode false. Cache mode true remains a reasonable mitigation for equal JAR sets, not a root-cause fix. Keep #10678 out of production as a blanket workaround until the exact retaining path and safe lifecycle contract are proven. No label or assignment change in this pass. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
