924060929 commented on PR #63366:
URL: https://github.com/apache/doris/pull/63366#issuecomment-4671961348

   ### Correction on the [P1] throw (AddLocalExchange)
   
   My earlier reply claimed the `throw` "never misfires" — that was wrong, and 
pushing it proved it. The PR's P0 / cloud_p0 runs surfaced **9 distinct 
legitimate patterns** where the throw fires but the plan is correct:
   
   | pattern | tests hit |
   |---|---|
   | `Serial UnionNode → non-serial TableFunctionNode` | unnest_join_list, 
unnest_where_list, salt_join, skew_join, correlated_scalar_subquery |
   | `Serial UnionNode → non-serial AggregationNode` | pull_up_predicate_set_op 
|
   | `Serial RecursiveCteScanNode → non-serial NLJ / Agg` | rec_cte, 
rec_cte_from_mysql_doc |
   | `Serial OlapScanNode → non-serial HashJoinNode` | 
test_crossjoin_inlineview_slot |
   
   In all of these the parent runs **effectively serial** at runtime and the BE 
reconciles `num_tasks` across the coupled pipelines, so no LocalExchange is 
required and the results are correct (they all pass with the `warn`). This 
validation compares FE-static `isSerialOperatorOnBe`, which can't model that 
runtime reconciliation — so it is **not a hard invariant**.
   
   I've reverted it to a `warn` (best-effort heuristic) with a comment 
explaining why a hard throw is unsafe here. Trying the throw was still 
worthwhile — it's exactly what surfaced these cases.
   
   ### Also fixed from the same CI run
   
   - **`force_passthrough` NPE:** `getConnectContext()` is null under the 
planner unit-test mocks, so the new `enableBroadcastJoinForcePassthrough` read 
NPE'd in `testHashJoinNodeBranches`. Made it null-safe (treat null context as 
"no force").
   - **clang-format:** `agg_operator_test.cpp` had a >100-col line (v16).
   
   (The remaining P0/cloud_p0 reds are pre-existing/flaky and unrelated to 
local shuffle — variant `element_at` logical-planner slot errors, a 
json_extract `IndexOutOfBounds`, a float→bigint cast overflow, a temp-table 
case, and one `SUCCESS vs RUNNING` timeout. The BE-UT red is a CI git-mirror 
checkout failure on the agent.)
   


-- 
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