Эта страница ещё не переведена: показана английская версия.

Logs: agents, syslog and SNMP traps

DumbMonit can keep the log lines of your machines and equipment next to their metrics, search them, and alert on patterns. Everything lands in the embedded VictoriaLogs and is searched on the admin-only Logs page with LogsQL.

Turn the log store on first

Settings → System → Log store. While it is off, received lines are dropped (and counted), and the Logs page says so.

What can feed it

Source	Stream (service)	How
DumbMonit itself	dumbmonit	Always, once the store is on.
Agents	agent	Per machine, from its device page.
Syslog	syslog	The built-in server (UDP and TCP).
SNMP traps	trap	The built-in receiver (v1, v2c).

Every line has level (error, warn, info, debug), host, origin (journald unit, file, event log channel or sender address) and, when the sender is one of your devices, target (its id) and target_name.

Agent log collection

Open an agent device and use Log collection. It is off until you switch it on; the agent applies the setting at its next report, with no restart.

• systemd journal (Linux): new entries from now on, filtered by priority.
• Followed files: absolute paths with * and ? in each component, for example /var/log/nginx/*.log. Rotation and truncation are followed; lines already in a file when collection starts are not replayed. Levels are guessed from the words error, warn, debug.
• Windows event log channels: System, Application… read through PowerShell every 15 seconds.
• Minimum level: lines below it are never sent.

Lines travel over the agent's usual channel (same server, same port, same token or certificate) in batches. If the server cannot be reached the agent keeps up to 8 MiB on disk (next to its binding secret) and sends them when it comes back; beyond that, the oldest are dropped. Set DUMBMONIT_AGENT_LOGS=off in the agent's environment to forbid collection on a machine whatever the server asks.

Warning

The paths come from the server, and the agent usually runs as root. Only administrators can set them, but treat the setting like the agent's command channel.

Syslog server

Off by default. Set DUMBMONIT_SYSLOG_ENABLED=true, publish the port in your Compose file (5514/udp and 5514/tcp, unprivileged because the container does not run as root), and point your equipment at it. RFC 3164 and RFC 5424 are understood, with newline-delimited or octet-counted TCP framing. The sender is attached to the device whose address is the source IP (or, failing that, whose name is the hostname in the message). Port 514 can be redirected to 5514 on the host or the router. Variables: see Configuration.

SNMP traps

Off by default (DUMBMONIT_TRAPS_ENABLED=true, port 1162/udp). SNMP v1 and v2c traps and v2c informs (acknowledged) are received; v3 is not supported and such messages are counted as refused. Set DUMBMONIT_TRAPS_COMMUNITIES to accept only your communities. Each trap becomes a line named after the trap (linkDown, coldStart…, or the OID) followed by its variable bindings, which are named after the built-in SNMP profiles when they know the OID.

Searching

The Logs page adds a Source (server, agents, syslog, SNMP traps, all) and a Device filter to the period, minimum level and LogsQL query. A device page has a Logs button that opens the page already filtered. The Receivers block shows what listens and what arrived.

Alerting on a pattern

Open Alert on this search below the search form: it creates a log pattern rule from the current source, level and query. The rule counts matching lines over a sliding window (1 minute to 24 hours) and fires when the count reaches your threshold, per device (one alert per target) or overall. It is a normal alert rule: severity, hold time, channels, silences, dependency suppression, history and backups all apply, and it can be edited on the Alerts page.

Examples (the query part):

• service:=syslog "Failed password" — failed logins reported through syslog;
• service:=trap level:in("error", "warn") — link-down and authentication traps, which is how you turn a trap into an alert;
• service:=agent origin:"file:/var/log/nginx/error.log" — nginx errors.

Rules only accept filters (no | pipes). A rule watching a log source that stays silent never fires: a count of zero is "no problem", so use >= with a threshold of at least 1.