Resolve Code Engine subnet pool id from a configured name

[Elena Peña Tapia (ElePT)]

Warning

Draft — pending confirmation in staging. The name→id lookup calls GET /projects/{id}/subnet_pools, which is present in the swagger SDK and modelled here, but I have not yet confirmed the target CE region actually serves it at the deployed API version. Before this leaves draft, run list_subnet_pools against a staging project (equivalently ibmcloud ce connectivity subnetpool list, the same endpoint) and confirm it returns id + name. Everything below is verified by unit tests and static checks only.

Summary

A Code Engine project can now be configured with a subnet pool name (subnet_pool_name) instead of its id. The IBM Cloud console shows a pool's name readily but not its id, so requiring the id added friction to every project onboarding.

When only a name is configured, the gateway resolves it to the id against the CE API on the first job submission for that project and caches the id back onto the CodeEngineProject row, so later jobs reuse it without another lookup. Supplying subnet_pool_id directly still works and skips resolution.

Because a pool name is not guaranteed unique within a project (the CE API only advises uniqueness; pools are identified by id), a name matching more than one pool fails the submission with an error naming the matching ids rather than guessing which pool to place the fleet on.

Changes

• Vendor SubnetPoolsApi.list_subnet_pools + the V2SubnetPool/V2SubnetPoolList models into the gateway ce_client (only list is needed; reuses the already-vendored pagination models).
• FleetHandler.resolve_subnet_pool_id — pages through the project's pools, matches by exact name, raises on zero or multiple matches.
• FleetsRunner.submit() — resolve-and-cache via a new _ensure_subnet_pool_id; subnet_pool_id made nullable and subnet_pool_name added (migration 0062).
• sync_ce_project — accepts either field, requires at least one, and invalidates a cached id only when the configured name changes (so re-sync doesn't wipe the cache every boot, given update_or_create overwrites all defaults).

Verification

• 158/158 tests pass across test_fleets_runner, fleets/ handler tests, and test_sync_ce_project (14 new). Covers: id-in-config skips resolution; name-only resolves once + caches; reuse without a second call; ambiguity/not-found raise; pagination; sync accept/reject and rename-invalidation.
• black ✅ · pylint 10.00/10 ✅ · import-linter ✅ · makemigrations --check clean ✅
• Not run: the live CE call (see warning above).

Out of scope (follow-up)

Resolving project_name→project_id and resource_group_name→resource_group_id. They're not symmetric with subnet pool: project_id is the row's primary key needed before a CE client exists (no lazy-at-submit hook), and resource_group_id lives in a different service (Resource Manager). Both should share the fail-loud-on-ambiguity contract used here.

[Marcelo Carneiro do Amaral (marceloamaral)]

Looks like a good feature to have as we discussed it before!