Dispatcher Service

The LibreNMS Dispatcher Service (librenms-service.py) schedules core LibreNMS tasks without using cron jobs. It does not replace the PHP scripts - only the cron entries that run them.

The Dispatcher handles scheduling for:

• Discovery
• Poller
• Services
• Alerting
• Billing
• Fast Ping
• Daily Maintenance

Setup

Install the Service

A systemd unit file is provided in misc/librenms.service. Update ExecStart and WorkingDirectory if LibreNMS is not installed in /opt/librenms.

cp /opt/librenms/misc/librenms.service /etc/systemd/system/librenms.service && systemctl enable --now librenms.service

Disable Cron Jobs

To prevent duplicate polling, disable existing cron-based pollers:

rm /etc/cron.d/librenms

Note: If using the librenms-scheduler cron (systemd timer), do not disable it.

Validate Setup

Let the poller run for a few minutes, then verify with:

./validate.php

Configuration

The dispatcher can be configured through:

• The Web UI
• Configuration defaults

Note: Settings are not applied until you restart the service.

Web UI

In the Web UI, go to Settings > Poller > Settings to manage node configurations. Nodes appear automatically after running for a few minutes.

Configuration Defaults

lnms config:set service_poller_workers 24
lnms config:set service_services_workers 8
lnms config:set service_discovery_workers 16
lnms config:set service_poller_frequency null
lnms config:set service_services_frequency 300
lnms config:set service_discovery_frequency 21600
lnms config:set service_billing_frequency 300
lnms config:set service_billing_calculate_frequency 60
lnms config:set service_poller_down_retry 60
lnms config:set service_loglevel INFO
lnms config:set service_update_frequency 86400

Memory Pressure

Optionally pause a dispatcher's intake when it is under memory pressure, to avoid OOM kills (for example during a thundering-herd startup of many workers). When usage reaches the configured percent the dispatcher stops taking new work, and resumes once it drops below (percent - 10). Unset (the default) disables it.

Usage is read from the process's cgroup memory limit when available (cgroup v2 or v1), otherwise from host memory. If usage cannot be determined the gate does nothing and polling behaves as normal.

This is set per dispatcher via an environment variable (there is no global config setting, as dispatchers sharing one LibreNMS install commonly have different memory limits):

export DISPATCHER_MEMORY_PRESSURE_PERCENT=85

Restrict Processing to Dispatcher

lnms config:set schedule_type.poller dispatcher
lnms config:set schedule_type.services dispatcher
lnms config:set schedule_type.discovery dispatcher
lnms config:set schedule_type.alerting dispatcher
lnms config:set schedule_type.billing dispatcher

Local Settings

Optional settings - define only in config.php:

$config['distributed_poller_name']  = php_uname('n');  // Unique poller name
$config['distributed_poller_group'] = 0;               // Poller group ID

Optimization

Worker Count

Dispatcher statistics appear in Settings > Poller > Poller.

Tip: Set the number of workers carefully. Too few threads delay the polling. Too many threads overload your hardware.

Keep Consumed Worker Seconds below Maximum Worker Seconds. If the two values are close and Devices Pending is 0, the poller is well tuned. If Devices Pending > 0 while Consumed Worker Seconds < Maximum Worker Seconds, consider Distributed Polling.

Maximum WS = workers × polling interval (default 300s)

Performance Tuning

See Performance Documentation for advanced tuning.

Distributed Polling

A single instance can poll up to 1,000+ devices, depending on latency and device responsiveness. If performance limits are reached, use Distributed Polling.

Fast Ping

The Fast Ping scheduler is disabled by default. If you use it, enable Fast Ping in the poller settings and set the following:

lnms config:set schedule_type.ping dispatcher

Systemd Watchdog Service

An alternate service file uses WatchdogSec to restart the dispatcher if it becomes unresponsive.

To install:

cp /opt/librenms/misc/librenms-watchdog.service /etc/systemd/system/librenms.service && systemctl enable --now librenms.service

Requires python3-systemd (or python-systemd on older systems). Without it, the service restarts every 30 seconds.