Alerts are mainly used for daily error response to improve system availability. KubeBlocks uses the open-source version of Prometheus to configure alert rules and multiple notification channels. The alert capability of KubeBlocks can meet the operation and maintenance requirements of production-level online clusters.
NOTE
The alert function is the same for all engines.
Alert rules
KubeBlocks uses the open-source version of Prometheus to meet the needs of various data products. These alert rules provide the best practice for cluster operation and maintenance, which further improve alert accuracy and reduce the probability of false negatives and false positives by experience-based smoothing windows, alert thresholds, alert levels, and alert indicators.
Taking PostgreSQL as an example, the alert rules have built-in common abnormal events, such as instance down, instance restart, slow query, connection amount, deadlock, and cache hit rate.
The following example shows PostgreSQL alert rules (refer to Prometheus for syntax). When the amount of active connections exceeds 80% of the threshold and lasts for 2 minutes, Prometheus triggers a warning alert and sends it to AlertManager.
alert: PostgreSQLTooManyConnections
expr: |
sum by (namespace,app_kubernetes_io_instance,pod) (pg_stat_activity_count{datname!~"template.*|postgres"})
> on(namespace,app_kubernetes_io_instance,pod)
(pg_settings_max_connections - pg_settings_superuser_reserved_connections) * 0.8
for: 2m
labels:
severity: warning
annotations:
summary: "PostgreSQL too many connections (> 80%)"
description: "PostgreSQL has too many connections and the value is {{ $value }}. (instance: {{ $labels.pod }})"
The alert message notification of KubeBlocks mainly adopts the AlertManager native capability. After receiving the Prometheus alert, KubeBlocks performs steps including deduplication, grouping, silence, suppression, and routing, and finally sends it to the corresponding notification channel.
