These are scripts I've put together for working with CESM on Cheyenne, as well as a few for working on Cheyenne generally.
I'm happy to listen to feature requests or other issues that you submit, but can't guarantee I'll have the capacity to address them. Feel free to fork this repo, make changes, and submit pull requests!
These are all written in Bash, so if that's your shell of choice you can call them just by invoking their name. Otherwise bash script_name.sh should work? I don't know, I'm allergic to other shells (there's a "shellfsh" joke in there somewhere). Feel free to email me or submit an issue with better instructions.
Some of these depend on others, so make sure they're on your PATH. E.g., for Bash, add this to your ~/.bash_profile:
PATH="$PATH:$HOME/scripts/other"
PATH="$PATH:$HOME/scripts/cesm"An example "chainspec" file for use in submit_chain.sh (see below).
My idea is to have a script I can call once to submit a job as well as all the jobs dependent on it. This is accomplished by calling submit_chain.sh with an argument pointing to a "chainspec" file (see chainspec.sh for an example). The chainspec files look basically like this, where runB1 and runB2 depend on state files generated by runA:
#!/bin/bash
list_cases=("runA"); list_deps=("")
list_cases+=("runB1"); list_deps+=("runA")
list_cases+=("runB2"); list_deps+=("runA")Some features:
- Can have jobs with RESUBMIT>0 (thanks to the
case.submit --resubmit-immediateflag). - Performs some sanity checks re: run order, run settings (
CONTINUE_RUN,RUN_TYPE,RUN_REF*) - Offers to remove unexpected restart files found in run directories.
Example usage:
cheyenne5:/glade/u/home/samrabin/cases_ctsm$ submit_chain.sh chain_20220726.05.sh
Case "chain_20220726.05.01"
Starts 1850-01-01
2 segments, each 5 days
Produces state for 1850-01-11
Case "chain_20220726.05.02"
This is a dependent run. Do you want to temporarily overwrite existing RUN_TYPE ("startup") with "branch"? y
Depends on case "chain_20220726.05.01"
Do you want to temporarily overwrite existing ref case (ifeuirne) with chain_20220726.05.01? y
Starts 1850-01-11
5 days
Produces state for 1850-01-16
Submitting case "chain_20220726.05.01"
Restart files already exist in run directory. Continue, Delete restart files, or Abort? d
rpointer files refer to 1850-01-11 instead of 1850-01-01. Continue, Delete rpointer files, or Abort? d
jobID_run: 5188445 5188447
jobID_archive: 5188446 5188448
Submitting case "chain_20220726.05.02"
Restart files already exist in run directory. Continue, Delete restart files, or Abort? d
rpointer files refer to 1850-01-16 instead of 1850-01-11. Continue, Delete rpointer files, or Abort? d
--prereq 5188448
jobID_run: 5188449
jobID_archive: 5188450
Big thanks to Dan Kennedy and Keith Oleson for inspiration and guidance; any annoyances with this script are on me.
Caveats:
- At the moment, I have it set up so that dependent runs must be branch runs; I'll eventually look into adding the ability for these to be hybrid as well. (Sooner if you submit an issue asking for this!)
- This depends on a CIME version that disables the check for
rpointerfiles incase.submitwhen it's called with--resubmit. This was introduced 2022-07-27 in CIME update 97d9f5a.
This script allows easy switching between different directories of a given case. For example, if I'm in the case's run directory, I can easily switch to its build, archive, or setup directory.
Note that I never call this script directly. Instead, I have added four lines to my ~/.bash_profile (after the addition of the script directories to my PATH):
alias cdc='. cd_custom.sh cases'
alias cda='. cd_custom.sh st_archive'
alias cdr='. cd_custom.sh run'
alias cdb='. cd_custom.sh bld'So, e.g., from the setup directory /glade/u/home/samrabin/cases_ctsm/casename/, I can do cdr to quickly switch to the run directory /glade/scratch/samrabin/cases_ctsm/casename/run/.
You'll need to customize the home_casedir variable to make this work for you.
Call this like copy_pes.sh /path/to/sourcecase from a case setup directory to copy the processor setup (i.e., XML variable NTASKS) from sourcecase to the case whose directory you're in.
From a case setup directory, call this script to create a directory (dir_links) with links to the run and archive directories. E.g.:
dir_links/
├── archive -> /glade/scratch/samrabin/archive/yield_perharv_smallville
└── run -> /glade/scratch/samrabin/yield_perharv_smallville/run
Sync all the log files from a case's archive and run directories to a new directory called run_logs in its setup directory (subdirectories run_logs/complete/ and run_logs/incomplete/, respectively). Call from setup directory.
I use this to avoid potentially-useful log files (e.g., from troubleshooting) getting deleted due to periodic /glade/scratch sweeps.
Call this from a case's run directory to follow the progress of the latest atm.log file. Add any argument to instead follow the latest file with that argument as a prefix; e.g., do tfncesm.sh cesm.log to follow the cesm.log file instead.
- When I say "case setup directory/folder," I mean the directory that you specify with
--casewhen callingcreate_newcaseorcreate_clone. - Many/most of these will probably break if you're in the habit of putting spaces in file or folder names. Sorry about that!