lots of stuff

2026-08-16 14:48:04 +02:00
parent c63db540e8
commit 49fa352476
45 changed files with 32520 additions and 93 deletions
@@ -0,0 +1,130 @@
+{
+  "nodes": [
+    {
+      "id": "internet",
+      "title": "Internet",
+      "subtitle": "public",
+      "group": "edge"
+    },
+    {
+      "id": "ingress",
+      "title": "Ingress",
+      "subtitle": "animeistrash",
+      "group": "edge",
+      "up": "up{job=\"node\",host=\"ingress\"}",
+      "stat": "100 - (avg(rate(node_cpu_seconds_total{job=\"node\",host=\"ingress\",mode=\"idle\"}[5m])) * 100)",
+      "stat_unit": "pct"
+    },
+    {
+      "id": "tunnel",
+      "title": "WireGuard",
+      "subtitle": "{{ wireguard_ingress_ip }} to {{ wireguard_beepi_ip }}",
+      "group": "edge",
+      "up": "max(wireguard_peer_up{peer=\"beepi\"})",
+      "stat": "max(wireguard_peer_handshake_age_seconds{peer=\"beepi\"})",
+      "stat_unit": "h"
+    },
+    {
+      "id": "beepi",
+      "title": "beepi",
+      "subtitle": "raspberry pi",
+      "group": "host",
+      "up": "up{job=\"node\",host=\"beepi\"}",
+      "stat": "100 - (avg(rate(node_cpu_seconds_total{job=\"node\",host=\"beepi\",mode=\"idle\"}[5m])) * 100)",
+      "stat_unit": "pct"
+    },
+{% for service in monitored_services %}
+    {
+      "id": "{{ service.name }}",
+      "title": "{{ service.title }}",
+      "subtitle": "{{ service.vhost | default(service.group) }}",
+      "group": "{{ service.group }}",
+      "enabled": "max(service_enabled{service=\"{{ service.name }}\"})",
+{% if service.probe is defined or service.tcp is defined %}
+      "up": "min(probe_success{service=\"{{ service.name }}\",vantage=\"local\"})",
+      "stat": "max(probe_duration_seconds{service=\"{{ service.name }}\",vantage=\"local\"})",
+      "stat_unit": "s"
+{% elif service.vhost is defined %}
+      "up": "min(probe_success{service=\"{{ service.name }}\",vantage=\"nginx\"})",
+      "stat": "max(probe_duration_seconds{service=\"{{ service.name }}\",vantage=\"nginx\"})",
+      "stat_unit": "s"
+{% else %}
+      "up": "min(service_units_active{service=\"{{ service.name }}\"} / service_units_total{service=\"{{ service.name }}\"})"
+{% endif %}
+    },
+{% endfor %}
+    {
+      "id": "raid",
+      "title": "RAID",
+      "subtitle": "{{ backup_raid_mount }}",
+      "group": "storage",
+      "up": "max(backup_raid_mounted)",
+      "stat": "max(node_filesystem_avail_bytes{mountpoint=\"{{ backup_raid_mount }}\"})",
+      "stat_unit": "bytes",
+      "fill": "1 - (max(node_filesystem_avail_bytes{mountpoint=\"{{ backup_raid_mount }}\"}) / max(node_filesystem_size_bytes{mountpoint=\"{{ backup_raid_mount }}\"}))"
+    },
+{% for job in backup_jobs %}
+{% if job.name != 'offsite' %}
+    {
+      "id": "backup-{{ job.name }}",
+      "title": "{{ job.title }}",
+      "subtitle": "backup job",
+      "group": "backup",
+      "up": "(max(backup_last_status{backup_job=\"{{ job.name }}\"}) == 1) and ((time() - max(backup_last_success_timestamp_seconds{backup_job=\"{{ job.name }}\"})) < {{ job.max_age_hours * 3600 }})",
+      "stat": "time() - max(backup_last_success_timestamp_seconds{backup_job=\"{{ job.name }}\"})",
+      "stat_unit": "h"
+    },
+{% endif %}
+{% endfor %}
+    {
+      "id": "offsite",
+      "title": "Offsite sync",
+      "subtitle": "daily 07:00",
+      "group": "storage",
+      "up": "(time() - max(backup_last_success_timestamp_seconds{backup_job=\"offsite\"})) < {{ 26 * 3600 }}",
+      "stat": "time() - max(backup_last_success_timestamp_seconds{backup_job=\"offsite\"})",
+      "stat_unit": "h"
+    },
+    {
+      "id": "hetzner",
+      "title": "Hetzner box",
+      "subtitle": "{{ hetzner_host | default('storage box') }}",
+      "group": "storage",
+      "up": "(max(hetzner_reachable) == 1) and (max(hetzner_remote_files{path=\"beepi\"}) > 0)",
+      "stat": "max(hetzner_remote_bytes{path=\"beepi\"})",
+      "stat_unit": "bytes",
+      "fill": "sum(hetzner_remote_bytes) / {{ hetzner_box_bytes }}"
+    },
+    {
+      "id": "uptimerobot",
+      "title": "External checks",
+      "subtitle": "uptimerobot",
+      "group": "edge",
+      "dashboard": "/d/beepi-uptimerobot",
+      "enabled": "max(uptimerobot_api_ok)",
+      "up": "min(uptimerobot_monitor_up and on (monitor) (uptimerobot_monitor_enabled == 1))",
+      "stat": "count(uptimerobot_monitor_enabled == 1)"
+    }
+  ],
+  "edges": [
+    {"source": "uptimerobot", "target": "internet", "label": "probes"},
+    {"source": "internet", "target": "ingress", "label": "public"},
+    {"source": "ingress", "target": "tunnel", "label": "dnat"},
+    {"source": "tunnel", "target": "beepi", "label": "wg0"},
+{% for service in monitored_services %}
+{% if service.vhost is defined and service.name != 'nginx' %}
+    {"source": "nginx", "target": "{{ service.name }}", "label": "proxy"},
+{% endif %}
+{% if service.vhost is not defined %}
+    {"source": "beepi", "target": "{{ service.name }}", "label": "{{ service.group }}"},
+{% endif %}
+{% endfor %}
+    {"source": "beepi", "target": "nginx", "label": "tls"},
+    {"source": "beepi", "target": "backup-daily", "label": "03:00"},
+    {"source": "backup-daily", "target": "backup-weekly", "label": "sun 05:00"},
+    {"source": "backup-daily", "target": "backup-restoretest", "label": "sun 06:00"},
+    {"source": "backup-daily", "target": "raid", "label": "archive"},
+    {"source": "raid", "target": "offsite", "label": "rsync"},
+    {"source": "offsite", "target": "hetzner", "label": "ssh 23"}
+  ]
+}