In my test run "ic-good-opt-on" and in general, we do not need to set
optimizer to "on". The optimizer is indeed set by default. In my run, you
will see in the test output that the test runner has determined that
optimizer is enabled by default:

https://github.com/edespino/cloudberry/actions/runs/11944371481/job/33295504090#step:15:117

============== checking optimizer status ==============
Optimizer enabled. Using optimizer answer files whenever possible

I now understand what you were trying to tell me. This
configuration "optimizer=on" is captured in the test results and I wasn't
setting it in my test run. Thank you for your review. I have started
another run with optimizer set to on.

Thank you buddy,
-=e

diff -I HINT: -I CONTEXT: -I GP_IGNORE: -U3
/__w/cloudberry/cloudberry/src/test/regress/expected/explain_optimizer.out
/__w/cloudberry/cloudberry/src/test/regress/results/explain.out
---
/__w/cloudberry/cloudberry/src/test/regress/expected/explain_optimizer.out
2024-11-20 16:59:55.210890622 -0800
+++ /__w/cloudberry/cloudberry/src/test/regress/results/explain.out
2024-11-20 16:59:55.230891152 -0800
@@ -395,7 +395,6 @@
          },                                                 +
          "Settings": {                                      +
              "Optimizer": "Pivotal Optimizer (GPORCA)",     +
-             "optimizer": "on",                             +
              "enable_parallel": "off",                      +
              "parallel_setup_cost": "0",                    +
              "parallel_tuple_cost": "0",                    +


On Wed, Nov 20, 2024 at 11:22 PM Ed Espino <esp...@apache.org> wrote:

> Thank you for sharing this. I am familiar with the previous Cloudberry
> workflow file and the use of it for "icw-orca-test".
>
> Again, I thought the default optimizer setting is "on". In this case, is
> there a need to even set it?
>
> Notice after creating the demo cluster, the optimizer is set to "on".
>
> https://github.com/edespino/cloudberry/actions/runs/11944371481/job/33295504090#step:14:247
>
> -=e
>
> On Wed, Nov 20, 2024 at 11:15 PM 杨江华 <yjhj...@gmail.com> wrote:
>
>> Hi, Ed
>> refer to cloudberry :
>>
>> https://github.com/apache/cloudberry/blob/1.5.4/.github/workflows/build.yml
>> ```c++
>> icw-orca-test:
>>     needs: build
>>     runs-on: [self-hosted, example]
>>     env:
>>       MAKE_TEST_COMMAND: "-k PGOPTIONS='-c optimizer=on'
>> installcheck-world"
>>       TEST_OS: "centos"
>>       DUMP_DB: "true"
>> ```
>> Here explicitly set PGOPTIONS='-c optimizer=on'.
>>
>>
>> Ed Espino <eesp...@gmail.com> 于2024年11月21日周四 15:00写道:
>>
>> > Please: you can call me Ed (艾德)
>> >
>> > I thought Orca (optimizer) was enabled by default. Is that not the case
>> > with Cloudberry?
>> >
>> > -=e
>> >
>> >
>> >
>> > On Wed, Nov 20, 2024 at 10:51 PM 杨江华 <yjhj...@gmail.com> wrote:
>> >
>> > > Dear Espino,
>> > >
>> > > I found that ic-good-opt-on pipeline failed to explain. You can set
>> > > PGOPTIONS='-c
>> > > optimizer=on' explicitly. Hope to solve this.
>> > >
>> > > Thank you for your time and support.
>> > >
>> > >
>> > > Ed Espino <esp...@apache.org> 于2024年11月21日周四 12:40写道:
>> > >
>> > > > I've identified a solution for running our previously disabled test
>> > > suites
>> > > > (
>> > > > cbdb_parallel and instr_in_shmem_verify) in the GitHub Actions CI
>> > > > environment. By increasing the container shared memory using the
>> > > > --shm-size=2gb parameter, these test suites are now passing
>> > consistently.
>> > > >
>> > > > I'm currently conducting additional testing to verify the stability
>> of
>> > > this
>> > > > solution. You can review the test results here:
>> > > >
>> > > >    -
>> https://github.com/edespino/cloudberry/actions/runs/11944371481
>> > > >    -
>> https://github.com/edespino/cloudberry/actions/runs/11943779438
>> > > >    -
>> https://github.com/edespino/cloudberry/actions/runs/11943000112
>> > > >
>> > > > However, I've noticed some inconsistencies in other test suites
>> during
>> > > this
>> > > > investigation. I would appreciate if you all could review these
>> results
>> > > and
>> > > > provide feedback on the increased shared memory solution and any
>> > insights
>> > > > into the inconsistent test results observed in other suites.
>> > > > Thank you,
>> > > > -=e
>> > > > --
>> > > > Ed Espino
>> > > > Apache Cloudberry (incubating) & MADlib
>> > > >
>> > >
>> >
>> >
>> > --
>> > Ed Espino
>> >
>>
>
>
> --
> Ed Espino
> Apache Cloudberry (incubating) & MADlib
>


-- 
Ed Espino
Apache Cloudberry (incubating) & MADlib

Reply via email to