Monitoring: fix dovecot rate charts, reduce storage to 60 days

Signed-off-by: Thilo Fromm <[email protected]>
t-lo · Dec 14, 2023 · 60352d7 · 60352d7
1 parent faa9e21
commit 60352d7
Show file tree

Hide file tree

Showing 4 changed files with 18 additions and 18 deletions.
diff --git a/grafana/dashboards/dovecot.json b/grafana/dashboards/dovecot.json
@@ -297,14 +297,14 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "rate(dovecot_auth_success_total[30s])*30",
+          "expr": "rate(dovecot_auth_success_total[1m])*60",
           "instant": false,
           "legendFormat": "{{user}}",
           "range": true,
           "refId": "A"
         }
       ],
-      "title": "Successful Authentications (1s rate, 30s avg)",
+      "title": "Successful Authentications (1s rate, 1m avg)",
       "transformations": [],
       "type": "timeseries"
     },
@@ -616,14 +616,14 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "rate(dovecot_auth_failures_total[30s])*30",
+          "expr": "rate(dovecot_auth_failures_total[1m])*60",
           "instant": false,
           "legendFormat": "{{user}}",
           "range": true,
           "refId": "A"
         }
       ],
-      "title": "Failed Authentications (1s rate, 30s avg)",
+      "title": "Failed Authentications (1s rate, 1m avg)",
       "transformations": [],
       "type": "timeseries"
     },

diff --git a/grafana/dashboards/main.json b/grafana/dashboards/main.json
@@ -705,14 +705,14 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "rate(dovecot_auth_success_count[30s])*30",
+          "expr": "rate(dovecot_auth_success_count[1m])*60",
           "instant": false,
           "legendFormat": "Avg. success per second",
           "range": true,
           "refId": "A"
         }
       ],
-      "title": "Successful Authentications (1s rate, 30s avg)",
+      "title": "Successful Authentications (1s rate, 1m avg)",
       "transformations": [],
       "type": "timeseries"
     },
@@ -1086,14 +1086,14 @@
           },
           "editorMode": "code",
           "exemplar": false,
-          "expr": "rate(dovecot_auth_failures_count[30s])*30",
+          "expr": "rate(dovecot_auth_failures_count[1m])*60",
           "instant": false,
           "legendFormat": "Avg. fails per second",
           "range": true,
           "refId": "A"
         }
       ],
-      "title": "Failed Authentications (1s rate, 30s avg)",
+      "title": "Failed Authentications (1s rate, 1m avg)",
       "transformations": [],
       "type": "timeseries"
     },
@@ -1611,7 +1611,7 @@
             "uid": "PBFA97CFB590B2093"
           },
           "editorMode": "code",
-          "expr": "rate(dovecot_sieve_action_count[30s])*30",
+          "expr": "rate(dovecot_sieve_action_count[1m])*60",
           "legendFormat": "Sieve actions",
           "range": true,
           "refId": "A"
@@ -1622,7 +1622,7 @@
             "uid": "PBFA97CFB590B2093"
           },
           "editorMode": "code",
-          "expr": "rate(dovecot_sieve_error_count[30s])*30",
+          "expr": "rate(dovecot_sieve_error_count[1m])*60",
           "hide": false,
           "legendFormat": "Sieve errors",
           "range": true,
@@ -1634,7 +1634,7 @@
             "uid": "PBFA97CFB590B2093"
           },
           "editorMode": "code",
-          "expr": "rate(dovecot_sieve_action_duration_seconds_sum[30s])*30",
+          "expr": "rate(dovecot_sieve_action_duration_seconds_sum[1m])*60",
           "hide": false,
           "legendFormat": "Sieve action duration",
           "range": true,

diff --git a/prometheus/prometheus.yaml b/prometheus/prometheus.yaml
@@ -1,24 +1,24 @@
 global:
-  scrape_interval:     5s
-  evaluation_interval: 5s
+  scrape_interval:     30s
+  evaluation_interval: 30s
 
 scrape_configs:
   - job_name: 'postfix'
-    scrape_interval: 5s
+    scrape_interval: 30s
     static_configs:
       - targets: ['mailserver:9154']
 
   - job_name: 'dovecot'
-    scrape_interval: 5s
+    scrape_interval: 30s
     static_configs:
       - targets: ['mailserver:9900']
 
   - job_name: 'fail2ban'
-    scrape_interval: 5s
+    scrape_interval: 30s
     static_configs:
       - targets: ['mailserver:9191']
 
   - job_name: 'pushgateway'
-    scrape_interval: 5s
+    scrape_interval: 30s
     static_configs:
       - targets: ['mailserver-prometheus-pushgateway:9091']
diff --git a/systemd/start_monitoring_prometheus.sh b/systemd/start_monitoring_prometheus.sh
@@ -13,5 +13,5 @@ exec docker run --rm --network mailserver-monitoring-internal \
            prom/prometheus:latest \
                 --config.file=/prometheus.yaml \
                 --storage.tsdb.path=/prometheus-data \
-                --storage.tsdb.retention.time=200d \
+                --storage.tsdb.retention.time=60d \
                 --web.enable-lifecycle