github-actions[bot] commented on code in PR #66228: URL: https://github.com/apache/doris/pull/66228#discussion_r3675111486
########## cloud/src/common/config.h: ########## @@ -279,8 +279,8 @@ CONF_mBool(enable_load_txn_status_check, "true"); CONF_mBool(enable_tablet_job_check, "true"); CONF_mBool(enable_recycle_delete_rowset_key_check, "true"); -CONF_mBool(enable_mark_delete_rowset_before_recycle, "true"); -CONF_mBool(enable_abort_txn_and_job_for_delete_rowset_before_recycle, "true"); +CONF_mBool(enable_mark_delete_rowset_before_recycle, "false"); Review Comment: [P1] Keep COMMITTED tmp rowsets fenced from recycling With both defaults false, `recycle_tmp_rowsets()` can delete and remove every tmp row of a transaction whose status is already `COMMITTED` but whose lazy conversion was interrupted (for example across restart). `do_recycle()` runs tmp recycling and `abort_timeout_txn()` concurrently; if recycling wins, `TxnLazyCommitTask::commit()` gets an empty scan, performs no partition conversion, and still calls `make_committed_txn_visible()`. The transaction becomes `VISIBLE` while its rowset files and metadata are gone, and its pending partition id can remain uncleared. No KV conflict spans object deletion, tmp-key removal, and that visibility transition; the ref-count key is removed only after data deletion. All nine modified lifecycle tests force true/true, so none exercises this default. Please at least preserve mark-before-delete and make deletion of COMMITTED tmp rows status-aware (advance or recheck the owner before physical deletion), with deterministic false/false recovery coverage. -- 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]
