zhangwenchao-123 commented on code in PR #1852:
URL: https://github.com/apache/cloudberry/pull/1852#discussion_r3629057480
##########
src/backend/gpopt/translate/CTranslatorDXLToPlStmt.cpp:
##########
@@ -5568,23 +5565,29 @@ CTranslatorDXLToPlStmt::TranslateDXLDirectDispatchInfo(
return segids_list;
}
- ULONG hash_code = GetDXLDatumGPDBHash(dxl_datum_array, pRTEHashFuncCal);
+ // Each datum array is one value combination of the distribution key
+ // (e.g. one array per IN-list constant). Dispatch to the union of the
+ // segments they hash to, like the planner does.
+ List *segids_list = NIL;
for (ULONG ul = 0; ul < length; ul++)
{
CDXLDatumArray *dispatch_identifier_datum_array =
(*dispatch_identifier_datum_arrays)[ul];
GPOS_ASSERT(0 < dispatch_identifier_datum_array->Size());
- ULONG hash_code_new = GetDXLDatumGPDBHash(
- dispatch_identifier_datum_array, pRTEHashFuncCal);
+ ULONG hash_code =
GetDXLDatumGPDBHash(dispatch_identifier_datum_array,
+
pRTEHashFuncCal);
- if (hash_code != hash_code_new)
- {
- // values don't hash to the same segment
- return NIL;
- }
+ segids_list = gpdb::LAppendUniqueInt(segids_list, hash_code);
Review Comment:
Maybe now is okay, as if we move `if (gpdb::ListLength(segids_list) ==
m_num_of_segments)` here, we will also add overhead in every iteration of the
foreach iteration, as a result, in some cases moving the judgement here is
better while worse in other cases.
--
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]
