search

EarthWorksOrg / EarthWorks

Other
3 stars 2 forks source link

Enable use of create_test workflow #45

Closed gdicker1 closed 3 months ago

gdicker1 commented 4 months ago

Add definitions of compsets and tests so that typical EarthWorks cases can be run as CIME tests instead.

Testing instructions (in this thread): https://github.com/EarthWorksOrg/EarthWorks/pull/45#issuecomment-2118204061

gdicker1 commented 4 months ago

TODOs:

gdicker1 commented 4 months ago

Some info about tests can be found in the CIME Documentation: https://esmci.github.io/cime/versions/master/html/users_guide/testing.html#testname-syntax

dazlich commented 3 months ago

I’ve got an outside opinion that it should be generally true.

gdicker1 commented 3 months ago

The various config_dt and ATM_NCPL values are set according to this table which comes from emails with Don Dazlich (and some input/context from Brian Dobbins).

Resolution ocn & ice dt (s) mpasa dt (s) ATM_NCPL (per day) Radiation Long & Short Wave
120km 1800.0 600 48 1hr (iradlw,iradsw=-1)
60km 900 300 96 1hr (iradlw,iradsw=-1)
30km 450 150 192 1hr (iradlw,iradsw=-1)
15km 240 80 360 40 min (iradlw,iradsw=10)
7.5km 120 40 720 20min (iradlw,iradsw=10
3.75km 60 20 1440 10min (iradlw,iradsw=10
gdicker1 commented 3 months ago

To test these changes:

  1. Clone this branch and run manage_externals
  2. Run create_test and point to the testlist_earthworks.xml file. ALSO ensure that you point to the EarthWorks input data and supply a suitable (EarthWorks) project/account key. Choose the test category from ew-pr, ew-ver, and ew-rel.
    1. NOTE: by default the builds will try to occur on the login node, running the command using the qcmd utility on Derecho is recommended to not violate fair use policies.
    2. NOTE: ew-ver and ew-rel can consume a lot of core hours. For now, please add --no-build to the create_test to ensure these tests only run their setup steps for now
  3. Once the cases have been submitted (or finished their appropriate step) you can check the status by going into the directory the tests were created in and running the cs.status.* script (the final part of the script name varies). You can grep for the "OVERALL" lines or for "FAIL" to limit output.

Example commands to run the ew-pr on Derecho and put the tests in <your_scratch>/ew_pr_tests

git clone -b feature/ew-cime-testing https://github.com/gdicker1/EarthWorks
cd EarthWorks && ./manage_externals/checkout_externals
qcmd -A <ew_proj_key> -- ./cime/scripts/create_test--xml-testlist cime_config/testlist_earthworks.xml \
 -i /glade/campaign/univ/ucsu0085/inputdata/ --project <ew_proj_key> \
 --test-root ${SCRATCH}/ew_pr_tests  --xml-category ew-pr
# This will block while builds and other steps finish
cd ${SCRATCH}/ew_pr_tests
./cs.status.*
gdicker1 commented 3 months ago

FULLYCOUPLED fails trying to find seaice/ocean data that doesn't exist in the campaign inputdata directory.

Judging by your command and comment I assume you mean the CESM input data on Campaign? Since I haven't figured out how to add EW as a model within CIME, we will keep getting that (CESM) location by default.

@dazlich, could you try adding -i /glade/campaign/univ/ucsu0085/inputdata and try again?

(Or go into the specific test, run ./xmlchange DIN_LOC_ROOT='/glade/campaign/univ/ucsu0085/inputdata' and then run ./test.submit)

dazlich commented 3 months ago

@gdicker1 Doh, of course - I am still getting up to speed on 'create_test'

dazlich commented 3 months ago
B2000-EW

I would have liked to called this CHAOS2000, CHAOS standing for Coupled Hexagonal Atmosphere Ocean and Seaice since that's what the coupled mpas components on a quasi-uniform grid essentially is. ;-)

gdicker1 commented 3 months ago

B2000-EW

I would have liked to called this CHAOS2000, CHAOS standing for Coupled Hexagonal Atmosphere Ocean and Seaice since that's what the coupled mpas components on a quasi-uniform grid essentially is. ;-)

Huh, I thought B2000 was already a compset (could still be, for all I know). I think CHAOS2000 is such a better name! If this is a vote, mine goes to CHAOS2000 instead :D

dazlich commented 3 months ago

Actually, I don’t see B2000 when I do a query_config, but other B-cases have CICE and either POP or MOM.

I choose the B2000-EW name since I set the bcs like GHG to cyclic 2000.

On May 20, 2024, at 4:01 PM, G. Dylan Dickerson @.***> wrote:

Caution: EXTERNAL Sender

B2000-EW

I would have liked to called this CHAOS2000, CHAOS standing for Coupled Hexagonal Atmosphere Ocean and Seaice since that's what the coupled mpas components on a quasi-uniform grid essentially is. ;-)

Huh, I thought B2000 was already a compset (could still be, for all I know). I think CHAOS2000 is such a better name! If this is a vote, mine goes to CHAOS2000 instead :D

— Reply to this email directly, view it on GitHubhttps://github.com/EarthWorksOrg/EarthWorks/pull/45#issuecomment-2121286372, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ADS3XWHEHB6DCHGW44A7WYLZDJXD3AVCNFSM6AAAAABHRB5AB2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMRRGI4DMMZXGI. You are receiving this because you were mentioned.Message ID: @.***>

gdicker1 commented 3 months ago

Just to add, I ran the entire ew-ver and ew-rel suites with the --no-build option and they all pass.

My intel/fullycoupled ew-pr ran out of time.

@dazlich, thank you. I'll update the FullyCoupled category, it'll only hurt the queue-length for these tests. When you have a sec, could you just make sure the sub-PRs make sense (i.e. pass visual inspection)?

dazlich commented 3 months ago

@gdicker1 I've had a look at the sub-PRs and they all look good to me!

areanddee commented 3 months ago

I think B2000 is a compset, rings a bell. I vote for CHAOS2000!

On Mon, May 20, 2024 at 4:01 PM G. Dylan Dickerson @.***> wrote:

B2000-EW

I would have liked to called this CHAOS2000, CHAOS standing for Coupled Hexagonal Atmosphere Ocean and Seaice since that's what the coupled mpas components on a quasi-uniform grid essentially is. ;-)

Huh, I thought B2000 was already a compset (could still be, for all I know). I think CHAOS2000 is such a better name! If this is a vote, mine goes to CHAOS2000 instead :D

— Reply to this email directly, view it on GitHub https://github.com/EarthWorksOrg/EarthWorks/pull/45#issuecomment-2121286372, or unsubscribe https://github.com/notifications/unsubscribe-auth/AZ2GWBNFBDYN7E76JXTSX43ZDJXD3AVCNFSM6AAAAABHRB5AB2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMRRGI4DMMZXGI . You are receiving this because your review was requested.Message ID: @.***>

dazlich commented 3 months ago

Big Boss rejected that name when I proposed it for our CSU model 20 years ago

On May 21, 2024, at 7:21 AM, areanddee @.***> wrote:

I think B2000 is a compset, rings a bell. I vote for CHAOS2000!