From d988ee64d3a850244035fbd81465b29e8c48cd68 Mon Sep 17 00:00:00 2001
From: Sheogorath <sheogorath@shivering-isles.com>
Date: Mon, 6 Nov 2023 12:54:33 +0100
Subject: [PATCH] feat(postgres): Add alert for WAL size
---
infrastructure/postgres/kustomization.yaml | 1 +
infrastructure/postgres/prometheusrules.yaml | 19 +++++++++++++++++++
2 files changed, 20 insertions(+)
create mode 100644 infrastructure/postgres/prometheusrules.yaml
diff --git a/infrastructure/postgres/kustomization.yaml b/infrastructure/postgres/kustomization.yaml
index a234bf501..6e94f812f 100644
--- a/infrastructure/postgres/kustomization.yaml
+++ b/infrastructure/postgres/kustomization.yaml
@@ -6,6 +6,7 @@ resources:
- repository.yaml
- release.yaml
- podmonitor.yaml
+ - prometheusrules.yaml
configMapGenerator:
- name: postgres-system-grafana-dashboards
files:
diff --git a/infrastructure/postgres/prometheusrules.yaml b/infrastructure/postgres/prometheusrules.yaml
new file mode 100644
index 000000000..0912f99d0
--- /dev/null
+++ b/infrastructure/postgres/prometheusrules.yaml
@@ -0,0 +1,19 @@
+---
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+ name: postgres-rules
+ namespace: postgres-system
+spec:
+ groups:
+ - name: postgres-clusters
+ rules:
+ - alert: PostgresHighWALUsage
+ annotations:
+ description: The WAL size of the postgres cluster exceeded 1GiB for more than 1 hour. This indicates a problem with either a replica or the WAL archiving process to S3.
+ summary: The WAL size of the postgres cluster exceeded 1GiB for more than 1 hour.
+ expr: max by (namespace, cluster_name)(pg_wal_size_bytes) > 2^30
+ for: 1h
+ labels:
+ issue: The WAL size of the postgres cluster exceeded 1GiB for more than 1 hour.
+ severity: critical
\ No newline at end of file
--
GitLab