provider ID is assigned to another device

[John A Meinel]

While debugging https://bugs.launchpad.net/juju/+bug/1945514

I came across this in the debug logs:

machine-0: 19:14:16 INFO juju.apiserver.common.networkingcommon processing link-layer devices for machine "57"
machine-0: 19:14:16 WARNING juju.apiserver.instancepoller not setting provider ID for device "enp5s0" to "7366"; it is assigned to another device
machine-0: 19:14:16 INFO juju.apiserver.common.networkingcommon processing link-layer devices for machine "74"
machine-0: 19:14:16 WARNING juju.apiserver.instancepoller not setting provider ID for device "enp5s0" to "7368"; it is assigned to another device
machine-0: 19:14:16 INFO juju.apiserver.common.networkingcommon processing link-layer devices for machine "82"
machine-0: 19:14:16 WARNING juju.apiserver.instancepoller not setting provider ID for device "enp5s0" to "7360"; it is assigned to another device
machine-0: 19:14:16 INFO juju.apiserver.common.networkingcommon processing link-layer devices for machine "79"
machine-0: 19:14:16 WARNING juju.apiserver.instancepoller not setting provider ID for device "enp5s0" to "7371"; it is assigned to another device
machine-0: 19:14:16 INFO juju.apiserver.common.networkingcommon processing link-layer devices for machine "73"
machine-0: 19:14:16 WARNING juju.apiserver.instancepoller not setting provider ID for device "enp5s0" to "7365"; it is assigned to another device
machine-0: 19:14:16 INFO juju.apiserver.common.networkingcommon processing link-layer devices for machine "76"
machine-0: 19:14:16 WARNING juju.apiserver.instancepoller not setting provider ID for device "enp5s0" to "7372"; it is assigned to another device

I don't know if this is related to: https://bugs.launchpad.net/juju/+bug/1898195

But it doesn't feel like something we should be seeing in the logs. Something is violating our expectations. (Either it is ok, and we shouldn't be warning, or it isn't and we need to be fixing something).

The logs and crashdump are available on the original bug.

If we find a simple way forward, or this takes too long to investigate, we can deprioritize this bug.

[Joseph Phillips (manadart)]

Is this recurrent?

The instance-poller merge logic does allow removal of a provider ID from an existing device, so the (unlikely) scenario in which one was on one and is not on another, we would expect eventual consistency.

If the provider ID is actually orphaned, I don't think the resolution is automatic. We did see this back in the 2.8 series, and we wrote an upgrade step for it:
https://github.com/juju/juju/pull/12132

So it seems there is a code path where we can get into this trouble still.