Jinja variables in rose stem cylc8 compatibility mode

Cylc Cylc 8 Migration
1

I’m trying to run UM (13.0) rose-stem and the suite is failing to validate or play due to undefined Jinja2 variables.
I haven’t used compatibility mode before, and I also haven’t run UM rose-stem before, so I don’t know if I’m seeing bugs, features, issues with my UM config or what else. The documentation does not answer all my questions!
Is rose-stem fully supported at cylc 8.2.1? or are there still things to be done?

From the documentation, I can see that the “defined on the fly at installation time” jinja2 variables are in opt/rose-suite-cylc-install.conf. It contains this:

# This file records CLI Options.

!opts=

[env]
# ROSE_ORIG_HOST set by cylc install.
ROSE_ORIG_HOST=gadi-login-06.gadi.nci.org.au

[jinja2]
HOST_SOURCE_UM="gadi-login-06.gadi.nci.org.au:/g/data/dp9/sjr548/um/r120067_um13_add_bom_cfg"
HOST_SOURCE_UM_BASE="gadi-login-06.gadi.nci.org.au:/g/data/dp9/sjr548/um/r120067_um13_add_bom_cfg"
RUN_NAMES=['fcm_make']
SITE="nci"
SOURCE_UM="/g/data/dp9/sjr548/um/r120067_um13_add_bom_cfg"
SOURCE_UM_BASE="/g/data/dp9/sjr548/um/r120067_um13_add_bom_cfg"
SOURCE_UM_MIRROR="fcm:um.xm/branches/dev/susanrennie/r120067_um13_add_bom_cfg@120401"
SOURCE_UM_REV=""

[jinja2:suite.rc]
# ROSE_ORIG_HOST set by cylc install.
ROSE_ORIG_HOST=gadi-login-06.gadi.nci.org.au
  • I’m not sure what is generating the entries under [jinja2], but they aren’t recognised when trying to play the workflow. This causes failure with cylc validate and cylc play.
  • If I use -S SITE='"nci"' with the rose-stem command then it shows up under [jinja2:suite.rc]
  • SITE is defined in global rose.conf in the cylc installation as [rose-stem]automatic-options=SITE=nci but I’m not sure if this is is the source of the SITE variable.
  • I don’t know where the rest of the stuff under [jinja2] is defined, but I guess somewhere in the UM rose-stem source code - blame my lack of knowledge on that one.
  • If I similarly add a variable to my own ~/.metomi/rose.conf, it does not get picked up at all - Should it?

Why do I get stuff under [jinja2] which probably should be under [jinja2:suite.rc]?

2

Okay, after some RTSC I found utilities.py in cylc-rose that does this:

    if templating != 'template variables':
        config['templating_detected'] = templating.replace(':suite.rc', '')
    else:
        config['templating_detected'] = templating

So now I know why I get jinja2 sans :suite.rc. But clearly that doesn’t explain how to get the workflow to recognise the jinja2 variables defined therein.

3

Why do I get stuff under [jinja2] which probably should be under [jinja2:suite.rc]?

Because there is a bug.

Thank you for bug-spotting and identifying exactly where and what the bug is.

As it happens we’ve already fixed this for the next release (this PR).

But clearly that doesn’t explain how to get the workflow to recognise the jinja2 variables defined therein.

The top level rose-suite.conf in the installed workflow is modified to include opts=(cylc-install) so that the variables saved in rose-suite-cylc-install.conf are made available when Cylc parses the flow.cylc file.

Your best bet for a practical fix is to change [jinja2:suite.rc] for [template variables] in your rose-suite.conf files.

If I similarly add a variable to my own ~/.metomi/rose.conf, it does not get picked up at all - Should it?

I don’t think so - have you found some documentation suggesting that you might expect this to happen?

4

As a workaround you can rename the jinja2:suite.rc section (Rose 2019/2) to template variables (Rose 2 only).

5

Thanks for the workaround!

My apologies for going to the bother of reading the source code (I thought I was looking at the most recent version yesterday but didn’t register the bugfix) and not searching through the issues for the problem.

I hadn’t seen the documentation mention user rose.conf with respect to rose-stem, but assumed any rose call might parse it. But I’m not very familiar with rose2 yet. A quick search of the documentation just now didn’t really clear things for me, unless the statement that rose-stem only wraps cylc install implies that things like parsing rose.conf are not included.

6

Spoke too soon.
If I use [template variables] in the rose.conf, then the suite fails to validate.

gadi-login-07:dp9 run6 $ cylc validate .
WARNING - Backward compatibility mode ON
FileParseError: Invalid line (line 1):
   {# The UM test suite is currently in a transitional state; the original   #}
(line numbers match 'cylc view -p')

It seems to be failing to parse all the jinja2 before validating?

If I use jinja2:suite.rc and manually define all the variables with -S when calling rose stem, then the suite validates fine

gadi-login-07:dp9 run5 $ cylc validate .
WARNING - Backward compatibility mode ON
INFO - nci
WARNING - deprecated items were automatically upgraded in "workflow definition"
WARNING -  * (8.0.0) [cylc]abort if any task fails - DELETED (OBSOLETE)
Valid for cylc-8.2.1

The only difference of note between run5 and run6 seems to be in the .conf files generated when installing.

7

Hi,

I think this error can be resolved (without changing [template variables]) by adding a #!Jinja2 shebang.

I.E. make #!Jinja2 the first line of the suite.rc file. This is what tells Cylc to run Jinja2 on the file.

I’ve spoken to the UM people, whilst Cylc 8 supports the rose-stem framework, the UM rose-stem workflow has not yet been upgraded to be Cylc 8 compatible on trunk. So sadly I think you’re going to find yourself fighting through the upgrade errors one at a time.

There is a branch here https://code.metoffice.gov.uk/trac/um/changeset?reponame=&new=119530%40main%2Fbranches%2Fdev%2Fjamesbruten%2Fvn13.3_um_cylc8_update&old=118802%40main%2Ftrunk but it isn’t fully functional just yet. It should validate and launch though. Note this isn’t on trunk, so it’s not supported and may change in the future.