github-actions[bot] commented on code in PR #66092:
URL: https://github.com/apache/doris/pull/66092#discussion_r3662813975


##########
fe/fe-core/src/main/java/org/apache/doris/cloud/transaction/CloudGlobalTransactionMgr.java:
##########
@@ -2787,6 +2787,12 @@ private void clearTxnLastSignature(long dbId, long 
txnId) {
      */
     private void notifyBesMakeTmpRsVisible(CommitTxnResponse commitTxnResponse,
                                            List<TabletCommitInfo> 
tabletCommitInfos) {
+        if (commitTxnResponse.getIsLazyCommit()

Review Comment:
   [P1] Make the skipped-promotion fallback safe for every read mode
   
   This return is safe with the default synchronous recovery settings, but the 
two read-side knobs are independent mutable configs. With 
`advance_txn_lazy_commit_during_reads=true` and 
`wait_txn_lazy_commit_during_reads=false`, `get_rowset` resets its KV 
transaction before submitting the lazy task, then falls through and 
dereferences `txn.get()==nullptr` while reading tablet stats. With advancement 
disabled, meta-service returns only the old rowsets; 
`CloudTablet::sync_rowsets` still returns OK below `query_version`, and point 
query then reads all local rowsets (or a versionless row-cache hit) without 
rechecking the requested version, so it can return the pre-commit value. 
Because this return deliberately withholds the best-effort promotion from the 
affected BE, correctness for an incomplete response depends on these recovery 
paths; please fail closed or make both fallback modes safe, and cover the two 
configurations end to end.



-- 
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]

Reply via email to