Incorrect early cycle earc dependency

[TravisElless-NOAA]

What is wrong?

PR #3202 added a chgres dependency to the earc job. However, since the chgres job is not included in the early cycle tasks, this dependency results in the early cycle earc job failing to launch.

What should have happened?

The early cycle earc task should not have a chgres dependency

What machines are impacted?

All or N/A

What global-workflow hash are you using?

01eeb24

Steps to reproduce

1. Create an experiment which runs the early cycle enkf (set EUPD_CYC="BOTH" in config.base)
2. Check dependencies for enkfgfs_eamn in xml file

Additional information

No response

Do you have a proposed solution?

The chgres dependency should only be included for the late cycle earc task.

[aerorahul]

@TravisElless-NOAA
Should we run the echgres job in the early cycle? I think we will need that for the GEFS initialization, no?

[aerorahul]

@TravisElless-NOAA
It is not difficult to add echgres dependency only to the RUN=enkfgdas:

diff --git i/workflow/rocoto/gfs_tasks.py w/workflow/rocoto/gfs_tasks.py
index 768512ba..08d260fe 100644
--- i/workflow/rocoto/gfs_tasks.py
+++ w/workflow/rocoto/gfs_tasks.py
@@ -2919,12 +2919,14 @@ class GFSTasks(Tasks):
         deps = []
         if 'enkfgdas' in self.run:
             dep_dict = {'type': 'metatask', 'name': f'{self.run}_epmn'}
-        else:
+            deps.append(rocoto.add_dependency(dep_dict))
+            dep_dict = {'type': 'task', 'name': f'{self.run}_echgres'}
+            deps.append(rocoto.add_dependency(dep_dict))
+            dependencies = rocoto.create_dependency(dep_condition='and', dep=deps)
+        else:  # early cycle enkf run (enkfgfs)
             dep_dict = {'type': 'task', 'name': f'{self.run}_esfc'}
-        deps.append(rocoto.add_dependency(dep_dict))
-        dep_dict = {'type': 'task', 'name': f'{self.run}_echgres'}
-        deps.append(rocoto.add_dependency(dep_dict))
-        dependencies = rocoto.create_dependency(dep_condition='and', dep=deps)
+            deps.append(rocoto.add_dependency(dep_dict))
+            dependencies = rocoto.create_dependency(dep=deps)

         earcenvars = self.envars.copy()
         earcenvars.append(rocoto.create_envar(name='ENSGRP', value='#grp#'))

Would you mind giving this patch a try in your test for early cycle?
The current CI tests do not have an early cycle EnKF test, so won't be able to test easily with current setups.
Thanks!

[TravisElless-NOAA]

My understanding of the chgres job is that it changes forecast field resolution for the current cycle's forecast. Since we are not running a forecast job in the early cycle there is nothing for this job to do. Any ensres forecast history files needed for a current analcalc job will still come from the chgres job in the previous late cycle.

This suggested change does indeed create the correct this dependency issue.