yiguolei opened a new pull request, #70691: URL: https://github.com/apache/doris/pull/70691
### Background Async result writer callbacks capture a weak task context and raw pointers to RuntimeState and the writer. Pipeline task diagnostics also access RuntimeState, whose lifetime is owned by the fragment. ### What problem does this PR solve? Issue Number: None Related PR: None PR Type: Bugfix Problem Summary: An async writer callback can start after cancellation releases its fragment. SCOPED_ATTACH_TASK previously dereferenced RuntimeState before validating the weak task context, so an expired callback could access freed memory. Lock the context before attachment, and move the instance-ID read in pipeline task diagnostics after its existing fragment lifetime check. ### Details - **Frequency:** Timing-dependent when an async writer remains queued during fragment teardown; no dynamic reproduction is claimed. - **Root cause:** RuntimeState was dereferenced before acquiring the context that protects its lifetime. - **Why existing tests missed it:** Async writer tests call process_block directly rather than exercising the queued callback attachment ordering. No reproducing test was identified for the diagnostic path. - **Fix:** Validate the captured weak task context before attaching the writer callback. Print the diagnostic instance ID only after obtaining a strong fragment reference and passing the finalized check. - **Other paths and branches:** The shared AsyncResultWriter entry covers its subclasses. No memory-reclamation or scheduler state-machine changes are included. - **Behavior change:** Expired writer callbacks return before accessing RuntimeState. Finalized or expired-fragment task summaries omit InstanceId; active summaries print it after the task summary. No protocol or storage-format changes. ### Testing - Repository clang-format v16 formatting/checking scripts passed on isolated copies of both modified files during development; the patches applied unchanged to the latest master. - git diff --cached --check passed on the PR branch. - Previously attempted ./run-be-ut.sh --run --filter=AsyncResultWriterTest.* -j 4 on macOS arm64; it stopped during Java environment validation because no JDK was installed. Tests did not compile or execute. - No unit tests added, as requested by the author. BE build, clang-tidy, and ASAN/TSAN reproduction remain unverified. ### Notes for Reviewers The diagnostic change removes an unguarded access, but a normal production UAF trigger for that particular diagnostic call has not been established. The async callback cancellation path is the primary fix. ### Release note Avoid a potential BE crash when an async result writer starts after its fragment has been cancelled and released. ### Check List (For Author) - Behavior changed: Yes (see Details) - Does this need documentation: No - Pre-submission status: Partially verified: formatting and static review completed; build and runtime validation remain outstanding. ### AI Disclosure - Tool: Codex - Generated: Code and PR description. - Human review: Pending. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
