[Bug] icw test fail on parallel_is_better case in gp_parallel

[smartyhero]

Cloudberry Database version

branch: main
commit id : b692345b

What happened

trigger build cloudberrydb, build success, but icw test is failed

case name: gp_parallel

log :

diff -I HINT: -I CONTEXT: -I GP_IGNORE: -U3 /code/gpdb_src/src/test/regress/expected/gp_parallel.out /code/gpdb_src/src/test/regress/results/gp_parallel.out
--- /code/gpdb_src/src/test/regress/expected/gp_parallel.out	2023-10-18 02:58:00.420410717 +0800
+++ /code/gpdb_src/src/test/regress/results/gp_parallel.out	2023-10-18 02:58:00.475410634 +0800
@@ -2036,9 +2036,11 @@
 begin;
 set local max_parallel_workers_per_gather = 8;
 select * from refresh_compare(true, false);
+NOTICE:  Non-parallel refresh duration=@ 3.283272 secs
+NOTICE:  Parallel refresh duration=@ 2.769643 secs
  parallel_is_better 
 --------------------
- t
+ f
 (1 row)
 
 select * from refresh_compare(false, false);

What you think should happen instead

No response

How to reproduce

use main branch (commit id b692345b) trigger ci/cd pipeline

Operating System

centos x86

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.

[github-actions]

Hey, @smartyhero welcome!🎊 Thanks for taking the time to point this out.🙌

[tuhaihe]

Hi @smartyhero, could you list the details here or ask for help in the internal Lark(飞书) group? Since people cannot access the code.hashdata.xyz in public.

[avamingli]

AFATK, this is caused by local machine performance for CI pipelines.
As a trade off, I once add a 0.8 coefficient to workaround.
But it still fails occasionally.

I will loosen that coefficient to pass regression.