Alertmanager state is not restored from remote storage if replicationFactor == 1

[callumj]

Describe the bug

Alertmanager state is not restored from remote storage if replicationFactor == 1: https://github.com/grafana/mimir/blob/main/pkg/alertmanager/state_replication.go#L205

I understand that running replicationFactor at 1 is a bad idea but for dev or staging environments it usually makes sense to keep things simple or cut down on costs.

I was wondering if this is intentional that we skip reading from storage in these cases?

To Reproduce

Steps to reproduce the behavior:

1. Run Alertmanager with replicationFactor of 1
2. Post an alert to AM
3. Kill the AM process
4. Observe state is not restored from storage

Expected behavior

State should be read from storage if no replication option is available

Environment

• Infrastructure: k8s
• Deployment tool: manual

Additional Context

[56quarters]

Looking at the change that introduced this, it seems like a bug to me too: 7896c43

[56quarters]

Actually, looks like it was a combination of the following PRs:

• Implement gRPC based initial state settling in alertmanager. cortexproject/cortex#3925
• Read alertmanager state from storage if peer settling fails. cortexproject/cortex#4021

I'll have something that should fix this up for review shortly.

[stevesg]

Good spot - the bug is that we should jump here if RF==1.