NATS registry caching services forever

[wkloucek]

Describe the bug

When using the NATS service registry, we can get quite quickly in a broken state.

Steps to reproduce

Steps to reproduce the behavior:

1. start the oCIS with NATS Helm chart example https://github.com/owncloud/ocis-charts/tree/master/deployments/ocis-nats
2. restart the gateway pods: kubectl -n ocis rollout restart deploy gateway

Expected behavior

Requests will fail for ~30 seconds. Afterwards everything is fine.

Actual behavior

Requests will fail forever. The proxy will log IPs of the gone gateway pods. ocis version will discovery the current gateway services when run inside any pod.

Additional context

oCIS 3.1.0-next.2

[micbar]

Severity? Blue Ticket?

[wkloucek]

Severity? Blue Ticket?

Not too pressing. Meanwhile I can use etcd but other projects may not. I guess ~ Feb 2024 would be nice.

[kobergj]

Needs an upstream fix in go-micro: https://github.com/go-micro/plugins/blob/v4/registry/nats/v1.2.1/v4/registry/nats/nats.go#L325 Change "create" to "put" (prolly we need a configuration for this) I'll file a PR

[kobergj]

go-micro PR is here: micro/plugins#120