Monitors and templates
A monitor evaluates data every minute and changes state. Start from a catalogue template, there are more than 90, or write your own with a live preview.
What the video shows
- The monitor list and their state.
- The "CPU saturated" monitor: query, thresholds, state of each group and history.
- New monitor: the template catalogue, by family.
States
| State | Meaning |
|---|---|
| OK | below thresholds |
| Warning | warning threshold crossed |
| Alert | critical threshold crossed |
| No data | no data over the window (no_data_s) |
A grouped monitor (by {host}) has one state per group. Each state change is recorded, notifies the monitor's channels and creates an event.
Types
| Type | What it evaluates |
|---|---|
metric | a metric query over a window (average, max, min, sum, last value) |
host | a host whose agent has sent nothing for the set delay |
service_check | the worst recent state of a check (systemd.unit, windows.service, tls.cert.expiry…) |
log | the number of lines of a search over a window, grouped by field or attribute |
apm | error rate, p95 latency or throughput of a service |
synthetic | N consecutive failures of a synthetic test |
anomaly | the share of points outside the usual bounds (rolling average, same slots of the last 7 days, or seasonal decomposition) |
forecast | the time before a curve crosses a threshold, by robust regression (for example, disk full within 48 h) |
outlier | the group that deviates from the others (median absolute deviation) |
composite | a logical expression over other monitors: 12 && (13 || !14) |
change | the change of a query compared with the same window earlier |
event | the number of events of a search (for example, SNMP traps) |
process | the number of processes with a given name on each host |
error_tracking | new or regressed issues of the Errors page |
slo | the remaining error budget or burn rate of an SLO |
The form shows a live preview: the groups, their state, the chart, and depending on the type the anomaly bounds or the forecast.
Thresholds and options
criticalandwarning, with an optional recovery threshold (critical_recovery,warning_recovery) to avoid flapping around a value.- Priority from P1 (critical) to P5 (information).
- Consecutive evaluations before alerting, and before returning to OK.
- Evaluation delay for late data.
- Host dependency (on by default): when a host stops answering, the other alerts of that host are held and do not notify. A monitor can also depend on other monitors.
- Flapping: beyond 5 changes in 30 minutes, a single "unstable" notification, then silence until things calm down.
- Renotification while the alert lasts, and escalations to other channels after a delay without acknowledgement.
- Mute until a date, or indefinitely.
The message
The notification message accepts variables and conditional blocks:
{{#is_alert}}{{name}} on {{host}}: {{value}} (threshold {{threshold}}).{{/is_alert}}
{{#is_warning}}{{host}} is close to the threshold: {{value}}.{{/is_warning}}
{{#is_recovery}}{{host}}: back to normal after {{duration}}.{{/is_recovery}}
Variables: {{name}}, {{group}}, {{host}}, {{value}}, {{last_value}}, {{state}}, {{priority}}, {{threshold}}, {{duration}}, {{url}}, {{tags.<key>}}, and {{eta}} for a forecast. Blocks: is_alert, is_warning, is_recovery, is_nodata. The mention @canal-<id> adds a channel for that message.
The template catalogue
Monitors › New monitor › From a template: more than 180 templates by family, each with its variables (targeted tags, thresholds, service). Template names and descriptions, and the monitors they create, are written in French for now; any monitor can be renamed.
| Family | Examples |
|---|---|
| System | CPU or memory saturated, disk I/O wait, disk almost full, disk full within 48 h, inodes, swap, abnormal load, host not reporting, reboot required |
| Services | systemd unit failed, Windows service stopped, Supervisor program failed |
| Security | certificate close to expiry, pending security updates, SSH brute-force attack, SSH unreachable, Wazuh daemon stopped, Wazuh agents disconnected, Suricata silent, packets dropped by Suricata, Suricata severity-1 alerts |
| Applications | HTTP 5xx errors, p95 latency, new application error |
| Databases, Message queues, Search | PostgreSQL, MySQL, Redis, Memcached, MongoDB, RabbitMQ, OpenSearch, PgBouncer, ProxySQL, ClickHouse, CouchDB, Couchbase |
| Web | PHP-FPM (listen queue, process cap reached), Varnish (sick backend, dropped sessions), Squid, Lighttpd, Gunicorn |
| Storage | ZooKeeper, Ceph cluster health, Ceph pool almost full, Btrfs degraded or out of unallocated space, hung NFS mount, NFS server stopped, NFS retransmissions |
| Directory and identity, DNS | OpenLDAP, Keycloak, PowerDNS and its Recursor (vulnerable version, SERVFAIL) |
| Repositories and registries, Logs | Nexus (read-only, health), Harbor (component down, disk), Fluentd (buffer full, output stuck) |
| Network | interface down, device unreachable |
| Postfix queue | |
| Synthetics | synthetic test failing |
| Kubernetes, Cloud, Virtualization | clusters, nodes, AWS, Azure, Google Cloud, OVHcloud, Scaleway, vSphere, Proxmox, Hyper-V, KVM |
| Generic collection | OpenMetrics target unreachable, series cap reached, Nagios plugin alerting, syslog reception failing, silent syslog device, dropped syslog lines, severe errors received by syslog |
| AI and GPU | Ollama, vLLM, LiteLLM, TGI, llama.cpp: unreachable, queue, latency; NVIDIA GPU too hot, saturated, memory full, faulty |
A new workspace receives six catalogue monitors, linked to an email channel to the account's address: host not reporting, disk almost full, memory saturated, systemd unit failed, Windows service stopped, certificate close to expiry.
Updated: 20 November 2026