github-actions[bot] commented on code in PR #64366:
URL: https://github.com/apache/doris/pull/64366#discussion_r3578523729
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/load/NereidsStreamLoadPlanner.java:
##########
@@ -324,20 +331,41 @@ public TPipelineFragmentParams plan(TUniqueId loadId, int
fragmentInstanceIdInde
queryOptions.setNewVersionUnixTimestamp(true);
Review Comment:
These globals are serialized into
`GroupCommitPlanner.execPlanFragmentParamsBytes` when a prepared group-commit
insert first builds the stream-load plan, but the prepared path reuses that
same `GroupCommitPlanner` on later executions when table/schema match. Only the
row bytes are rebuilt, so later executions keep the first execution's
timestamp/timezone in the cached fragment. That matters for omitted columns
with defaults such as `DEFAULT CURRENT_TIMESTAMP`:
`NereidsLoadPlanInfoCollector` serializes those default expressions into the
file scan params, and BE evaluates them against the fragment `TQueryGlobals`. A
prepared group-commit insert that omits the timestamp column can therefore load
later rows with the first execution's time or timezone. Please refresh/rebuild
the cached group-commit fragment globals per execution, or invalidate the
cached planner when the execution statement time/timezone changes.
##########
be/src/service/point_query_executor.cpp:
##########
@@ -345,10 +354,7 @@ Status PointQueryExecutor::init(const
PTabletKeyLookupRequest* request,
*_tablet->tablet_schema(), 1));
}
}
- // Set timezone from request for functions like from_unixtime()
- if (request->has_time_zone() && !request->time_zone().empty()) {
- _reusable->runtime_state()->set_timezone(request->time_zone());
- }
Review Comment:
This still installs the request globals after a cache-miss
`Reusable::init()` has opened the output expressions. `Reusable::init()` calls
`VExpr::open(_output_exprs_ctxs, _runtime_state.get())`, and `VExpr::open()`
materializes all-constant expressions into `_constant_col` at fragment-local
scope. So a prepared point query such as `SELECT /*+
SET_VAR(debug_skip_fold_constant=true,time_zone='+00:00') */ from_unixtime(20)
FROM t WHERE k=?` can serialize the correct timezone/timestamp in
`PTabletKeyLookupRequest`, but the cache-miss open already folded
`from_unixtime(20)` using the default/stale runtime state; later
`update_runtime_state()` cannot change the cached constant. Please apply the
request timezone/timestamp before `Reusable::init()` prepares/opens the exprs,
or pass the globals into `Reusable::init()`, and add coverage with an
all-constant timezone-sensitive output.
--
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]
