Having come from running regional ocean models a lot, then trying rAM3, I started from the main rAM3 guide.. Here are a few issues I ran into that might be useful feedback for the rAM3 developers
Following the linked rose/cylc instructions, I initially had trouble because these instructions didn’t tell me to follow this initialisation step in the NCI’s cylc7 instructions. Without running the one line bash script from this step, I get the error: no such identity: /home/149/ab8992/.ssh/id_rsa-rose-cylc-gadi: No such file or directory Suggestion:
Update instructions to include this extra step OR modify the /site/nci-gadi configuration to work with the current set of instructions
My default project was v45, an old project with no more compute. When running ocean models, I’m used to setting the project in the configuration file. In the ancil suite (u-bu503), there’s an option for NCI_PROJECT in the rose-suite.conf file. Setting this did successfully submit some jobs with the correct project, but others still read my default project. This is an easy fix (just change your default project) but does suggest that the overwriting of the PROJECT field isn’t working as expected. This issue will likely catch other MOM6 users out who might expect this functionality! Suggestion:
Add a line to the instructions so that users know that their default project will be read.
Remove the NCI_PROJECT= "" line from the rose-suite.conf, add a comment explaining the limitation, or ensure that it does indeed set the project correctly
The Lismore example states that OSTIA is optional, but does not tell the user how to turn it off. This may seem trivial to an experienced user, but since the instructions said it was optional, I assumed the suite would run if I just skipped the OAS step! It does not, unless you set USE_OSTIA = false in the rose-suite.conf file. Suggestion:
Default to OSTIA being false
Add a line in the optional OAS step of the instructions to tell the user how to turn it on
For me, turning OSTIA on led to issues. I’ve tried running the OAS multiple times from scratch on a fresh clone (or whatever it’s called in UM’s upcycled retro git). Following the logs, the log/job/1/ostia_netcdf_to_pp/01/job.out file has print statements promising me that heaps of files have been, quote: Written to /scratch/gb02/ab8992/OSTIA_ANCIL/20220227T0000Z_ostia_sst.pp. However, when running the RNS, it fails because some of these files don’t exist. @mlipson has suggested they could be getting deleted in a housekeeping step, @heidi has said that maybe it’s intermittent and I just need to retry until all of the files write successfully. I haven’t been able to diagnose the problem, so am just leaving a note of it here.
The final issue I had was a really dumb one. I hadn’t yet been added to the BARRA project… rookie mistake I know. However, the error that was spat out was very cryptic, and didn’t seem to me to be a ‘file not found’ / ‘permission denied’ error, so it didn’t occur to me that this was the problem. Looking at the code,
# Read in the surface temperature data from the archive
BARRA_FIELDN = 'ts'
indir = os.path.join(BARRA_DIR, '1hr',BARRA_FIELDN, 'latest')
barra_files = glob(os.path.join(indir, BARRA_FIELDN + '*' + yyyy + mm + '*nc'))
barra_fname = indir + '/' + barra_files[0].split('/')[-1]
I'd suggest adding the following after the `barra_files = glob(...)` line:
if not barra_files:
raise ImportError("BARRA files not found. This most likely means you're not a member of the ob53 group.")
The initialisation step was included in the instructions when I wrote them but they seem to have been deleted. I will raise this with the documentation team.
There is definitely some confusion about the ability to run with different projects in the RNS. The answer is that you need to run with the project set in the gadi environment. It is not possible to switch between projects as you can with other model runs.
The default setting of OSTIA true/false has been discussed at length and it was decided to set it to true because it is a new feature that we wanted users to be able to use. It is unfortunately not a clear-cut easy choice to make because of different sets of user needs,
I am not sure what is going on with your OAS. We may need to schedule a time to chat (perhaps after AMOS – apologies I have some work I need to attend to)
And yes, the error message when a project has not been joined could be made clearer.
The initialisation step not most likely not in the generic rose/cylc instructions because it is only required when running ram3 (as far as I know). A bit awkward but thanks for pointing this out. @heidi do you want to chase that issue up?
Yes you’re right it is in the detailed guide section - perhaps just needs to be also put in the quick guide. I’d assumed that following the quick guide (which sends you to the rose/cylc page for most of the setup), would end up doing the same thing as the detailed guide so I didn’t see it the first time I tried!
But you do specifically refer new users to the “detailed guide”, so that’s on me I guess
No - it is not on you, it is a lot. The problem is that there are nuanced differences between the models. Hopefully the docs team can come up with a solution.
I’ll just add that I’ve seen Ash’s issue of not being in all the required groups come up with two other new users, and they were also stuck on that for some time.
It’s hard for users to keep track of the dozen or so groups they need to be a part of to run successfully, as they are all approved at different times.
Perhaps a simple check could be included when running a suite that checks that the users groups are compliant, and a clear error message if not?
Thanks for pointing that out. It is difficult to anticipate what causes new users problems when those type of issues just don’t trip you up – hopefully you know what I mean.
For the moment I will definitely emphasise that in training and yes we could have a simple “check you group memberships” type script added to the starter tools.
I will look into it.
Thanks again
1 Like
clairecarouge
(Claire Carouge, ACCESS-NRI Land Modelling Team Lead)
10
@ashjbarnes , I’ll remove the help tag since you aren’t quite asking for help and instead providing feedback and ways to improve rAM3 configuration and documentation. Let me know if you mind and we can discuss more.
