From: Tom Hughes <tom@compton.nu>
Date: Thu, 16 Feb 2023 19:03:53 +0000 (+0000)
Subject: Reduce sensitivity of postgres replication alarms
X-Git-Url: https://git.openstreetmap.org./chef.git/commitdiff_plain/67c0ac7256b64455f3f95cc727485bbe05a10515

Reduce sensitivity of postgres replication alarms
---

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 367e07650..1736da60c 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -99,8 +99,8 @@ groups:
   - name: database
     rules:
       - alert: postgres replication delay
-        expr: pg_replication_lag_seconds > 5
-        for: 5m
+        expr: pg_replication_lag_seconds > 30
+        for: 15m
         labels:
           alertgroup: database
         annotations:
@@ -489,8 +489,8 @@ groups:
         labels:
           alertgroup: "{{ $labels.instance }}"
       - alert: postgresql replication delay
-        expr: pg_replication_lag_seconds > 5
-        for: 1m
+        expr: pg_replication_lag_seconds > 30
+        for: 15m
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations: