[Bug] Parallel query plan changes, plan is unstable

[oracleloyall]

Apache Cloudberry version

Latest main branch

What happened

      ======================================================================
      DIFF FILE: /workspace/database/src/test/regress/regression.diffs
      ----------------------------------------------------------------------
      diff -I HINT: -I CONTEXT: -I GP_IGNORE: -U3 /workspace/database/src/test/regress/expected/orca_parallel.out /workspace/database/src/test/regress/results/orca_parallel.out
--- /workspace/database/src/test/regress/expected/orca_parallel.out	2026-03-02 14:57:53.726035231 +0800
+++ /workspace/database/src/test/regress/results/orca_parallel.out	2026-03-02 14:57:53.754037247 +0800
@@ -690,18 +690,30 @@
 ___________
  Gather Motion 6:1  (slice1; segments: 6)
    Output: pidx1.a, pidx1.b
-   ->  Parallel Hash Semi Join
+   ->  Result
          Output: pidx1.a, pidx1.b
-         Hash Cond: (pidx1.a = pidx2.c)
-         ->  Parallel Index Scan using pidx1_a_idx on orca_parallel.pidx1
-               Output: pidx1.a, pidx1.b
-               Index Cond: (pidx1.a < 500)
-         ->  Parallel Hash
-               Output: pidx2.c
-               ->  Parallel Index Only Scan using pidx2_c_idx on orca_parallel.pidx2
-                     Output: pidx2.c
-                     Index Cond: (pidx2.c < 500)
-GP_IGNORE:(15 rows)
+         ->  Unique
+               Output: pidx1.a, pidx1.b, (RowIdExpr)
+               Group Key: (RowIdExpr)
+               ->  Sort
+                     Output: pidx1.a, pidx1.b, (RowIdExpr)
+                     Sort Key: (RowIdExpr)
+                     ->  Redistribute Motion 6:6  (slice2; segments: 6)
+                           Output: pidx1.a, pidx1.b, (RowIdExpr)
+                           Hash Key: (RowIdExpr)
+                           Hash Module: 3
+                           ->  Parallel Hash Join
+                                 Output: pidx1.a, pidx1.b, (RowIdExpr)
+                                 Hash Cond: (pidx2.c = pidx1.a)
+                                 ->  Parallel Index Only Scan using pidx2_c_idx on orca_parallel.pidx2
+                                       Output: pidx2.c
+                                       Index Cond: (pidx2.c < 500)
+                                 ->  Parallel Hash
+                                       Output: pidx1.a, pidx1.b, (RowIdExpr)
+                                       ->  Parallel Index Scan using pidx1_a_idx on orca_parallel.pidx1
+                                             Output: pidx1.a, pidx1.b, RowIdExpr
+                                             Index Cond: (pidx1.a < 500)
+GP_IGNORE:(27 rows)

What you think should happen instead

No response

How to reproduce

CI test error, instability occurs

Operating System

aarch64

Anything else

No response

Are you willing to submit PR?

• Yes, I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct.