Support bundles creating prior to online update not accessible

[askfongjojo]

I've noticed all rack2 bundles show up with failed status. To confirm whether they are indeed tied to a particular nexus node, I made a new one during update execution.

$ oxide bundle view --bundle-id cd00c20a-4949-4025-963c-7b9c3efc0f31
{
  "id": "cd00c20a-4949-4025-963c-7b9c3efc0f31",
  "reason_for_creation": "Created by external API",
  "state": "active",
  "time_created": "2025-10-18T05:59:09.652031Z"
}

I was able to inspect its contents.

After the update process got pass nexus handoff, the bundle status changed to failed. Inspect or download is no longer allowed:

$ oxide bundle view --bundle-id cd00c20a-4949-4025-963c-7b9c3efc0f31
{
  "id": "cd00c20a-4949-4025-963c-7b9c3efc0f31",
  "reason_for_creation": "Created by external API",
  "reason_for_failure": "Nexus managing this bundle no longer exists",
  "state": "failed",
  "time_created": "2025-10-18T05:59:09.652031Z"
}

$ oxide bundle inspect --id cd00c20a-4949-4025-963c-7b9c3efc0f31
downloading support bundle index cd00c20a-4949-4025-963c-7b9c3efc0f31: Error Response: status: 400 Bad Request; headers: {"content-type": "application/json", "x-request-id": "492c1313-501d-4500-8838-a6a2d324c5f4", "content-length": "151", "date": "Sat, 18 Oct 2025 07:02:52 GMT"}; value: <stream>

[smklein]

There are two conditions by which a support bundle can be marked failed:

1. The dataset we picked to manage the bundle has been expunged (FAILURE_REASON_NO_DATASET)
2. The Nexus we picked to collect and manage the bundle has been expunged and is ready for cleanup (FAILURE_REASON_NO_NEXUS)

I think it's very possible that we're hitting the second condition here, as Nexuses are added + expunged, rather than being updated in-place. This would happen when "all the new nexuses" come online, see bundles assigned to the set of old nexuses, realize that those old nexuses are expunged, and consequently mark the corresponding bundles as failed.

This seems like it's marking bundles failed too conservatively - I'll see if I can sketch out a way to avoid this failure mode, and keep active bundles alive across updates.

[askfongjojo]

Right, I think it's acceptable for bundles in collecting state to fail after switching to a new set of nexus (operator just have to make a new request). But I'd expect bundles in active state to remain accessible.