nevzheng commented on issue #13515: URL: https://github.com/apache/gravitino/issues/13515#issuecomment-5879698680
**The bounded-fork stack from #13529 is in review with real evidence; a further threading spike is now in progress.** **Status:** #13530 (bounded adaptive forks) → #13553 (share one container per backend, replacing per-fork containers) are both open PRs, stacked in that order. #13553 has real before/after timing on hardware matched to `gcp-arc-runners-large` (8 vCPU/31 GB), satisfying this epic's evidence guardrail — container count confirmed 2→1 per backend, no test regressions, no deadlocks. Recommendation is to merge. **Why it matters:** this closes the "generalize safe container and fixture reuse" line item above with measured proof, not a projection. **What's in progress now:** a config-flag-gated spike (`-PcoreDatabaseThreaded=true`, default off) investigating whether intra-JVM thread parallelism can replace process-fork parallelism for these same tests — a further resource-overhead reduction if it holds up. Root-cause research (cited, file:line-level) already ruled out any Postgres/JDBC-level blocker; the constraint is 4 JVM-wide static singletons in the storage backend, which are being scoped per-backend-instance in an isolated worktree. This respects the epic's own guardrail — no PR yet, evidence first, decision gate after benchmarking. **Go deeper:** #13530, #13553. -- 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]
