Rangsh commented on issue #12266: URL: https://github.com/apache/seatunnel/issues/12266#issuecomment-5883868936
Thanks @DanielLeens @SEZ9 — fully aligned on both notes. I will keep the 4 GiB JMH fork budget and the earlier full-WAL reload OOM (`34317023975` on `432bdb3d9`) as supporting acceptance context only. I will **not** treat the local `-Xmx2g` single-key smoke (`5bbc304b2`), or a heap budget alone, as proof of selective loading / no-retention. The functional gate stays as you both stated: 1. **Phase A** (first, after #12173 is healthy and merged): prove the keyed file-storage contract with deterministic `FileMapStore` / WAL-reader tests (filler keys, overwrite/tombstone cases, no unrelated keys, no whole-map retention), plus a constrained-heap diagnostic on the same `initialStoredJobCount=1000` shape as like-for-like supporting evidence. The SPI default full-load-and-filter path will not be what Phase B relies on. 2. **Phase B** (only after Phase A is merged and shown selective): widen the durability sample to the full batch **outside** the measured `SingleShot` path, with an explicit tear-down-only assertion in the initial revision. Phase B will reference the merged Phase A change. On the parent PR: #12173 head remains at `7948e2640` (no further pushes unless a re-review asks for changes). Code review items F1–F6 are closed there; the remaining gate is the stale `CHANGES_REQUESTED` docs review (already addressed in `9a6077c58`) plus clearing the blocked merge state. I will **not** start a storage or benchmark implementation PR for this issue, and will not stack work on that branch, until #12173 is healthy and merged. No label/assignee changes from my side. Thanks again for the clear gate. -- 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]
