I want to configure ruler alert which will check the connectivity of ruler with backend azure blob storage.

[shreekumartatte]

Will below alerts rule will help me ?

alert: RulerStorageConnectionErrors
expr: rate(mimir_ruler_storage_request_errors_total[5m]) > 0
for: 3m
labels:
severity: critical
annotations:
summary: "Ruler storage API connection errors"
description: "The ruler is unable to reach Azure Blob Storage or is experiencing connectivity/DNS/authentication issues."

[apoorva-01]

That rule won't fire, because mimir_ruler_storage_request_errors_total isn't a metric Mimir exposes. The ruler's object storage client is instrumented through the Thanos objstore wrapper, same as the compactor and store-gateway, so the metric you actually want is:

sum(rate(thanos_objstore_bucket_operation_failures_total{component="ruler-storage"}[5m])) > 0

The component label is the piece to get right. Mimir builds the ruler's bucket client with the name "ruler-storage", and that name becomes the component label on every objstore metric (pkg/storage/bucket/client.go wraps the registerer with {"component": name}). So component="ruler-storage" is your ruler-to-Azure operations, component="compactor" is the compactor's, and so on.

One thing worth deciding: > 0 on a rate() is pretty twitchy. A single transient blip (Azure throttling, one DNS hiccup) will page you. If you just want "the ruler can't reach storage," a small threshold over a longer window is calmer, something like:

alert: RulerStorageConnectionErrors
expr: sum(rate(thanos_objstore_bucket_operation_failures_total{component="ruler-storage"}[5m])) > 0.01
for: 10m

Check the exact label values against your own /metrics first (namespace/cluster labels depend on your setup), but thanos_objstore_bucket_operation_failures_total with the ruler-storage component is the right signal for ruler storage connectivity.

[s3onghyun]

That rule will not fire, because mimir_ruler_storage_request_errors_total is not a metric Mimir exposes.

Mimir's ruler reaches object storage through the Thanos objstore layer, which emits thanos_objstore_bucket_operation_failures_total. That metric carries a bucket label set to the name of the storage client, and the ruler's client is named ruler-storage (in pkg/ruler/storage.go the bucket client is created with that name). So the signal you want is:

sum(rate(thanos_objstore_bucket_operation_failures_total{bucket="ruler-storage"}[5m])) > 0

A couple of refinements:

• A bare rate(...) > 0 is twitchy and will page on a single transient error. Consider a for: window, or averaging over a few minutes.
• thanos_objstore_bucket_operation_duration_seconds (same bucket label) is useful if you also want to alert on latency/degradation rather than only hard failures.