github-actions[bot] commented on code in PR #66360:
URL: https://github.com/apache/doris/pull/66360#discussion_r3699769294
##########
be/src/format_v2/table_reader.cpp:
##########
@@ -765,6 +765,104 @@ Status TableReader::_build_table_filters_from_conjuncts()
{
return Status::OK();
}
+namespace {
+
+bool same_scan_projection(const LocalColumnIndex& lhs, const LocalColumnIndex&
rhs) {
+ if (lhs.index != rhs.index || lhs.project_all_children !=
rhs.project_all_children ||
+ lhs.children.size() != rhs.children.size()) {
+ return false;
+ }
+ for (size_t index = 0; index < lhs.children.size(); ++index) {
+ if (!same_scan_projection(lhs.children[index], rhs.children[index])) {
+ return false;
+ }
+ }
+ return true;
+}
+
+const LocalColumnIndex* find_scan_projection(const FileScanRequest& request,
+ LocalColumnId column_id) {
+ const auto find_by_id = [column_id](const std::vector<LocalColumnIndex>&
projections) {
+ return std::ranges::find_if(projections, [column_id](const
LocalColumnIndex& projection) {
+ return projection.column_id() == column_id;
+ });
+ };
+ auto it = find_by_id(request.predicate_columns);
+ if (it != request.predicate_columns.end()) {
+ return &*it;
+ }
+ it = find_by_id(request.non_predicate_columns);
+ return it == request.non_predicate_columns.end() ? nullptr : &*it;
+}
+
+bool same_physical_scan_layout(const FileScanRequest& lhs, const
FileScanRequest& rhs) {
+ if (lhs.local_positions != rhs.local_positions) {
+ return false;
+ }
+ for (const auto& [column_id, _] : lhs.local_positions) {
+ const auto* lhs_projection = find_scan_projection(lhs, column_id);
+ const auto* rhs_projection = find_scan_projection(rhs, column_id);
+ if (lhs_projection == nullptr || rhs_projection == nullptr ||
+ !same_scan_projection(*lhs_projection, *rhs_projection)) {
+ return false;
+ }
+ }
+ return true;
+}
+
+} // namespace
+
+Status TableReader::refresh_conjuncts(VExprContextSPtrs conjuncts) {
Review Comment:
[P2] Keep late-runtime-filter rebuild work in the owning profile hierarchy.
This new path rebuilds TableReader filters/mapping and opens localized
expressions, then Parquet validates the refreshed projection, but it runs
before `TableReader::get_block()` without TableReader/FileReader scopes. As a
result that latency is charged only to FileScannerV2 and the
TableReader/Parquet subtree cannot explain a slow late-RF refresh. Please add
non-overlapping refresh scopes at the TableReader and concrete reader/JNI
boundaries (taking care not to double-count hybrid wrappers), plus a profile
regression for an open-reader refresh.
--
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]
