Fix #813: continue create pipeline on config-save HTTP 500

[dacasti2mx]

Related Issue(s)

Fixes #813

Related Collection Role

• cisco.nac_dc_vxlan.validate
• cisco.nac_dc_vxlan.dtc.create
• cisco.nac_dc_vxlan.dtc.deploy
• cisco.nac_dc_vxlan.dtc.remove
• other

Related Data Model Element

• vxlan.topology

Proposed Changes

On a pre-provisioned fabric the switches can still be reloading or
unreachable when the create pipeline runs the intermediate config-save.
NDFC then returns HTTP 500. 0.7.2 logged this and continued; 0.8.0 aborts
at inventory_config_save, so the rest of the fabric intent is never
applied. This is a regression.

Root cause: pipeline_base._config_save already meant to treat a 500 as
non-fatal, but the check could never fire. fabric_deploy_manager discarded
the NDFC response on a failed config-save and returned failed=True with no
msg, so the RETURN_CODE was lost before the caller could look at it.

Fix:

• fabric_config_save() returns the NDFC response.
• manage_fabrics() propagates it into results['msg'] when config-save
  fails, so the RETURN_CODE survives.
• _config_save() treats RETURN_CODE == 500 as non-fatal and emits a clear
  warning (full detail under -v). Any other failure — 400, auth, or a
  malformed/missing response — still aborts the run. This is stricter and
  safer than the old 0.7.2 rescue, which swallowed every error.

Only the create pipeline registers _config_save (create_resources.yml);
remove_resources.yml does not, and the deploy role uses operation: all,
so deploy and remove behavior are unchanged.

Test Notes

The collection has no unit-test harness today, and CI runs only build,
sanity, and lint — so a proper unit test needs new infrastructure. To avoid
blocking the 0.8.1 fix on that project-structure decision, tests are
deferred and tracked in a follow-up: #818.

Verified locally with a temporary mock harness (not committed) that drives
the real _config_save:

• HTTP 500 -> run continues, warning emitted
• HTTP 400 -> run aborts (fatal)
• missing or malformed msg -> run aborts (fatal)
• HTTP 200 -> success

Deferred unit-test matrix (in the follow-up):

• 500 -> manager returns failed=True with the response intact
• 500 -> _config_save returns non-fatal
• 400 -> fatal
• missing/malformed msg -> fatal
• 200 -> ok

Cisco Nexus Dashboard Version

4.1 (per issue #813)

Checklist

• Latest commit is rebased from develop with merge conflicts resolved
• New or updates to documentation has been made accordingly
• Assigned the proper reviewers

[dacasti2mx]

Update: the 500 bypass is now gated on the data model.

Before, this PR treated any config-save HTTP 500 as non-fatal — so a real 500 on a healthy fabric would also be swallowed.

The latest commit only tolerates the 500 when the data model actually has pre-provisioned switches (poap.preprovision) — i.e. the case where NDFC returns 500 because the switches aren't reachable yet. A 500 on a fabric without pre-provisioned switches stays fatal (as do 400 / auth / malformed).

Verified live against a pre-provisioned fabric: config-save returns 500, the run continues, and create completes 21/21. The new check was also validated deterministically (preprov → tolerate, non-preprov / empty → fatal).

[ccoueffe]

LGTM

[mthurstocisco]

Tested and works for a full pre provisioned fabric for me

[juburnet]

Approve