diff options
Diffstat (limited to 'Prometheus/alerts')
| -rw-r--r-- | Prometheus/alerts/lysergic/blackbox.yml | 94 | ||||
| -rw-r--r-- | Prometheus/alerts/lysergic/nginx_exporters.yml | 13 | ||||
| -rw-r--r-- | Prometheus/alerts/lysergic/node_exporters.yml | 13 | ||||
| -rw-r--r-- | Prometheus/alerts/lysergic/wireguard_exporters.yml | 23 | ||||
| -rw-r--r-- | Prometheus/alerts/tripsit/blackbox.yml | 21 | ||||
| -rw-r--r-- | Prometheus/alerts/tripsit/node_exporters.yml | 13 |
6 files changed, 177 insertions, 0 deletions
diff --git a/Prometheus/alerts/lysergic/blackbox.yml b/Prometheus/alerts/lysergic/blackbox.yml new file mode 100644 index 0000000..49a553d --- /dev/null +++ b/Prometheus/alerts/lysergic/blackbox.yml @@ -0,0 +1,94 @@ +groups: +- name: lysergic-blackbox_exporters + rules: + - alert: LIBERTACASA-WEB-DOWN + expr: probe_success{instance=~"https://liberta.casa|https://www.lysergic.dev"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: XKEK-WEB-DOWN + expr: probe_success{instance=~"https://xkek.net|https://kekx.net"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: JITSI-WEB-DOWN + expr: probe_success{instance=~"https://meet.lysergic.dev"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: PASTA-WEB-DOWN + expr: probe_success{instance=~"https://pasta.lysergic.dev"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: GITEA-WEB-DOWN + expr: probe_success{instance=~"https://git.com.de"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: CGIT-WEB-DOWN + expr: probe_success{instance=~"https://git.casa"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: ELEMENT-WEB-DOWN + expr: probe_success{instance=~"https://element.liberta.casa"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: ETHERPAD-WEB-DOWN + expr: probe_success{instance=~"https://pad.hugz.io"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: SEARX-WEB-DOWN + expr: probe_success{instance=~"https://searx.xkek.net"} == 0 + for: 120s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. + - alert: YACY-WEB-DOWN + expr: probe_success{instance=~"https://yacy.xkek.net"} == 0 + for: 30s + labels: + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 30 seconds. Node seems down. +/etc/prometheus/alerts/lysergic/blackbox.yml lines 29-93/93 (END) diff --git a/Prometheus/alerts/lysergic/nginx_exporters.yml b/Prometheus/alerts/lysergic/nginx_exporters.yml new file mode 100644 index 0000000..8a3777d --- /dev/null +++ b/Prometheus/alerts/lysergic/nginx_exporters.yml @@ -0,0 +1,13 @@ +groups: +- name: lysergic-nginx_exporters + rules: + - alert: lysergic-nginx-down + expr: nginx_up{project="LYSERGIC", job="nginx_exporters"} == 0 + for: 1m + labels: + job: nginx_exporters + severity: WARNING + project: LYSERGIC + annotations: + title: nginx {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 1 minute. nginx seems down. diff --git a/Prometheus/alerts/lysergic/node_exporters.yml b/Prometheus/alerts/lysergic/node_exporters.yml new file mode 100644 index 0000000..392b3b7 --- /dev/null +++ b/Prometheus/alerts/lysergic/node_exporters.yml @@ -0,0 +1,13 @@ +groups: +- name: lysergic-node_exporters + rules: + - alert: lysergic-node-down + expr: up{project="LYSERGIC", job="node_exporters"} == 0 + for: 1m + labels: + job: node_exporters + severity: warning + project: LYSERGIC + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 1 minute. Node seems down. diff --git a/Prometheus/alerts/lysergic/wireguard_exporters.yml b/Prometheus/alerts/lysergic/wireguard_exporters.yml new file mode 100644 index 0000000..87fecb6 --- /dev/null +++ b/Prometheus/alerts/lysergic/wireguard_exporters.yml @@ -0,0 +1,23 @@ +groups: +- name: lysergic-wireguard_exporters + rules: + - alert: mercury-wireguard-down + expr: wireguard_peer_last_handshake_seconds{device="wg0", instance="localhost:9586", job="wireguard-mercury", public_key="$WG0_PUBKEY"} - time() <- 125 + for: 10s + labels: + job: wireguard-mercury + severity: CRITICAL + project: LYSERGIC + annotations: + title: WireGuard {{ $labels.instance }} is down + description: 'Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 2 minutes. Uplink to Psyched seems down. Affected services: Global infrastructure monitoring. Other outages might no longer be tracked. Urgent investigation is advised.' + - alert: lysergic-wireguard-down + expr: wireguard_peer_last_handshake_seconds{project="LYSERGIC",device!="wg1"} - time() <- 130 + for: 10s + labels: + job: wireguard-lysergic + severity: CRITICAL + project: LYSERGIC + annotations: + title: WireGuard Incident + description: 'No handshakes for {{ $labels.device }} in {{ $labels.job }} on {{ $labels.instance}} have been exchanged for an extended amount of time. Critical services could be affected.' diff --git a/Prometheus/alerts/tripsit/blackbox.yml b/Prometheus/alerts/tripsit/blackbox.yml new file mode 100644 index 0000000..0390718 --- /dev/null +++ b/Prometheus/alerts/tripsit/blackbox.yml @@ -0,0 +1,21 @@ +groups: +- name: tripsit-blackbox_exporters + rules: + - alert: TRIPSIT.ME-WEB-INCIDENT + expr: probe_success{instance=~"https://tripsit.me|https://chat.tripsit.me|https://chat.tripsit.me/chat|https://drugs.tripsit.me|https://benzo.tripsit.me|https://dxm.tripsit.me|https://combo.tripsit.me|https://wiki.tripsit.me|https://tripbot.tripsit.me"} == 0 + for: 15s + labels: + severity: critical + project: TRIPSIT + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 15 seconds. Node seems down. + - alert: TRIPSIT.DEV-WEB-INCIDENT + expr: probe_success{instance=~"https://mail.tripsit.dev|https://dopamine.tripsit.dev"} == 0 + for: 1m + labels: + severity: warning + project: TRIPSIT + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 1 minute. Node seems down. diff --git a/Prometheus/alerts/tripsit/node_exporters.yml b/Prometheus/alerts/tripsit/node_exporters.yml new file mode 100644 index 0000000..ee82367 --- /dev/null +++ b/Prometheus/alerts/tripsit/node_exporters.yml @@ -0,0 +1,13 @@ +groups: +- name: tripsit-node_exporters + rules: + - alert: TRIPSIT-NODE-INCIDENT + expr: up{project="TRIPSIT", job="node_exporters"} == 0 + for: 30s + labels: + severity: warning + job: node_exporters + project: TRIPSIT + annotations: + title: Node {{ $labels.instance }} is down + description: Failed to scrape {{ $labels.job }} on {{ $labels.instance}} for more than 1 minute. Node seems down. |