Merge pull request grafana/cortex-jsonnet#406 from grafana/alert-on-consul-failures

Added CortexFailingToTalkToConsul alert

Author: pracucci

jsonnet/mimir-mixin/alerts/alerts.libsonnet

@@ -235,6 +235,27 @@
             |||,
           },
         },
+        {
+          alert: 'CortexKVStoreFailure',
+          expr: |||
+            (
+              sum by(%s, pod, status_code, kv_name) (rate(cortex_kv_request_duration_seconds_count{status_code!~"2.+"}[1m]))
+              /
+              sum by(%s, pod, status_code, kv_name) (rate(cortex_kv_request_duration_seconds_count[1m]))
+            )
+            # We want to get alerted only in case there's a constant failure.
+            == 1
+          ||| % [$._config.alert_aggregation_labels, $._config.alert_aggregation_labels],
+          'for': '5m',
+          labels: {
+            severity: 'warning',
+          },
+          annotations: {
+            message: |||
+              Cortex {{ $labels.pod }} in  %(alert_aggregation_variables)s is failing to talk to the KV store {{ $labels.kv_name }}.
+            ||| % $._config,
+          },
+        },
         {
           alert: 'CortexMemoryMapAreasTooHigh',
           expr: |||

jsonnet/mimir-mixin/docs/playbooks.md

@@ -734,6 +734,20 @@ How to **investigate**:
 - Ensure there's no pod `NotReady` (the number of ready containers should match the total number of containers, eg. `1/1` or `2/2`)
 - Run `kubectl -n <namespace> describe statefulset <name>` or `kubectl -n <namespace> describe deployment <name>` and look at "Pod Status" and "Events" to get more information
 
+### CortexKVStoreFailure
+
+This alert fires if a Cortex instance is failing to run any operation on a KV store (eg. consul or etcd).
+
+How it **works**:
+- Consul is typically used to store the hash ring state.
+- Etcd is typically used to store by the HA tracker (distributor) to deduplicate samples.
+- If an instance is failing operations on the **hash ring**, either the instance can't update the heartbeat in the ring or is failing to receive ring updates.
+- If an instance is failing operations on the **HA tracker** backend, either the instance can't update the authoritative replica or is failing to receive updates.
+
+How to **investigate**:
+- Ensure Consul/Etcd is up and running.
+- Investigate the logs of the affected instance to find the specific error occurring when talking to Consul/Etcd.
+
 ## Cortex routes by path
 
 **Write path**: