Add MimirIngesterStuckProcessingRecordsFromKafka alert

Signed-off-by: Marco Pracucci <[email protected]>

docs/sources/mimir/manage/mimir-runbooks/_index.md

@@ -1427,6 +1427,22 @@ How to **investigate**:
 
 - Check ingester logs to see why requests are failing, and troubleshoot based on that.
 
+### MimirIngesterStuckProcessingRecordsFromKafka
+
+This alert fires when an ingester has successfully fetched records from Kafka but it's not processing them at all.
+
+How it **works**:
+
+- Ingester reads records from Kafka, and processes them locally. Processing means unmarshalling the data and handling write requests stored in records.
+- Fetched records, containing write requests, are expected to be processed ingesting the write requests data into the ingester.
+- This alert fires if no processing is occurring at all, like if the processing is stuck (e.g. a deadlock in ingester).
+
+How to **investigate**:
+
+- Take goroutine profile of the ingester and check if there's any routine calling `pushToStorage` and what's it state:
+  - If the call exists and it's waiting on a lock then there may be a deadlock.
+  - If the call doesn't exist then it could either mean processing is not stuck (false positive) or the `pushToStorage` wasn't called at all, and so you should investigate the callers in the code.
+
 ### MimirIngesterFailsEnforceStrongConsistencyOnReadPath
 
 This alert fires when too many read-requests with strong consistency are failing.

operations/helm/tests/metamonitoring-values-generated/mimir-distributed/templates/metamonitoring/mixin-alerts.yaml

@@ -988,6 +988,19 @@ spec:
             for: 5m
             labels:
               severity: critical
+          - alert: MimirIngesterStuckProcessingRecordsFromKafka
+            annotations:
+              message: Mimir {{ $labels.pod }} in {{ $labels.cluster }}/{{ $labels.namespace }} is stuck processing write requests from Kafka.
+              runbook_url: https://grafana.com/docs/mimir/latest/operators-guide/mimir-runbooks/#mimiringesterstuckprocessingrecordsfromkafka
+            expr: |
+              # Alert if the reader is not processing any records, but there buffered records to process in the Kafka client.
+              # NOTE: the cortex_ingest_storage_reader_buffered_fetch_records_total metric is a gauge showing the current number of buffered records.
+              (sum by (cluster, namespace, pod) (rate(cortex_ingest_storage_reader_records_total[5m])) == 0)
+              and
+              (sum by (cluster, namespace, pod) (cortex_ingest_storage_reader_buffered_fetch_records_total) > 0)
+            for: 5m
+            labels:
+              severity: critical
           - alert: MimirIngesterFailsEnforceStrongConsistencyOnReadPath
             annotations:
               message: Mimir {{ $labels.pod }} in {{ $labels.cluster }}/{{ $labels.namespace }} fails to enforce strong-consistency on read-path.

operations/mimir-mixin-compiled-baremetal/alerts.yaml

@@ -962,6 +962,19 @@ groups:
           for: 5m
           labels:
             severity: critical
+        - alert: MimirIngesterStuckProcessingRecordsFromKafka
+          annotations:
+            message: Mimir {{ $labels.instance }} in {{ $labels.cluster }}/{{ $labels.namespace }} is stuck processing write requests from Kafka.
+            runbook_url: https://grafana.com/docs/mimir/latest/operators-guide/mimir-runbooks/#mimiringesterstuckprocessingrecordsfromkafka
+          expr: |
+            # Alert if the reader is not processing any records, but there buffered records to process in the Kafka client.
+            (sum by (cluster, namespace, instance) (rate(cortex_ingest_storage_reader_records_total[5m])) == 0)
+            and
+            # NOTE: the cortex_ingest_storage_reader_buffered_fetch_records_total metric is a gauge showing the current number of buffered records.
+            (sum by (cluster, namespace, instance) (cortex_ingest_storage_reader_buffered_fetch_records_total) > 0)
+          for: 5m
+          labels:
+            severity: critical
         - alert: MimirIngesterFailsEnforceStrongConsistencyOnReadPath
           annotations:
             message: Mimir {{ $labels.instance }} in {{ $labels.cluster }}/{{ $labels.namespace }} fails to enforce strong-consistency on read-path.

operations/mimir-mixin-compiled/alerts.yaml

@@ -976,6 +976,19 @@ groups:
           for: 5m
           labels:
             severity: critical
+        - alert: MimirIngesterStuckProcessingRecordsFromKafka
+          annotations:
+            message: Mimir {{ $labels.pod }} in {{ $labels.cluster }}/{{ $labels.namespace }} is stuck processing write requests from Kafka.
+            runbook_url: https://grafana.com/docs/mimir/latest/operators-guide/mimir-runbooks/#mimiringesterstuckprocessingrecordsfromkafka
+          expr: |
+            # Alert if the reader is not processing any records, but there buffered records to process in the Kafka client.
+            # NOTE: the cortex_ingest_storage_reader_buffered_fetch_records_total metric is a gauge showing the current number of buffered records.
+            (sum by (cluster, namespace, pod) (rate(cortex_ingest_storage_reader_records_total[5m])) == 0)
+            and
+            (sum by (cluster, namespace, pod) (cortex_ingest_storage_reader_buffered_fetch_records_total) > 0)
+          for: 5m
+          labels:
+            severity: critical
         - alert: MimirIngesterFailsEnforceStrongConsistencyOnReadPath
           annotations:
             message: Mimir {{ $labels.pod }} in {{ $labels.cluster }}/{{ $labels.namespace }} fails to enforce strong-consistency on read-path.

operations/mimir-mixin/alerts/ingest-storage.libsonnet

@@ -96,6 +96,7 @@
           },
         },
 
+        // Alert firing if an ingester is failing to read from Kafka.
         {
           alert: $.alertName('IngesterFailsToProcessRecordsFromKafka'),
           'for': '5m',
@@ -110,6 +111,25 @@
           },
         },
 
+        // Alert firing is an ingester is reading from Kafka, there are buffered records to process, but processing is stuck.
+        {
+          alert: $.alertName('IngesterStuckProcessingRecordsFromKafka'),
+          'for': '5m',
+          expr: |||
+            # Alert if the reader is not processing any records, but there buffered records to process in the Kafka client.
+            (sum by (%(alert_aggregation_labels)s, %(per_instance_label)s) (rate(cortex_ingest_storage_reader_records_total[5m])) == 0)
+            and
+            # NOTE: the cortex_ingest_storage_reader_buffered_fetch_records_total metric is a gauge showing the current number of buffered records.
+            (sum by (%(alert_aggregation_labels)s, %(per_instance_label)s) (cortex_ingest_storage_reader_buffered_fetch_records_total) > 0)
+          ||| % $._config,
+          labels: {
+            severity: 'critical',
+          },
+          annotations: {
+            message: '%(product)s {{ $labels.%(per_instance_label)s }} in %(alert_aggregation_variables)s is stuck processing write requests from Kafka.' % $._config,
+          },
+        },
+
         {
           alert: $.alertName('IngesterFailsEnforceStrongConsistencyOnReadPath'),
           'for': '5m',