oom-messages-exporter 20230920

Collects OOM Killer messages from /var/log/messages and represents the data for Prometheus.

• OOM Messages Exporter
• Examples
  • OOM inside kubernetes containers
  • OOM on bare metal servers or VMs
• How to run
  • Examples
  • Alerting rules
  • Production
  • Config
• Dependencies

OOM Messages Exporter

An exporter for providing OOM Killer's kills in the Prometheus format. The exporter is compatible with containers runtimes and could be useful to investigate OOM Killer activity inside containers, servers or vms.

Examples

The main metrics are 'oom_messages_exporter_last_killed_pid' and 'oom_messages_exporter_last_killed_time' those provide values of the last pid and the last time of killed processes since the exporter has started. You can configure Alertmanager on the ..._time metric (value > now - last Xh) or on a change of the ..._pid metric by expression delta(_pid[xH]) != 0 function. Also you can combine both metric to get better alerts experience.

Provided labels are:

• command - a proc title of a killed process;
• namespace (opt, containers runtimes only)
• pod (opt, containers runtimes only)
• container (opt, containers runtimes only)

OOM inside kubernetes containers

oom_messages_exporter_last_killed_pid{command="php-fpm: pool w",container="php-fpm",namespace="oldwebsite",pod="a-legacy-app"} 235362.0
oom_messages_exporter_last_killed_time{command="framework",container="nodejs",namespace="lovely-project",pod="awesome-backend"} 1.69484585513779e+09

OOM on bare metal servers or VMs

For this case labels about containers runtime should be as 'None'

oom_messages_exporter_last_killed_pid{command="node /opt/www/p",container="None",namespace="None",pod="None"} 24190.0
oom_messages_exporter_last_killed_time{command="node /opt/www/p",container="None",namespace="None",pod="None"} 1.6948459852485702e+09

How to run

Examples

Collecting OOM Killer messages about processes in an operating system:

# ./examples/exporter-container.sh

Collecting OOM Killer messages inside containers:

# ./examples/exporter-systemd.sh

Output (exporter-container.sh):

$ curl -s http://localhost:9001/ | egrep -v '^#' | grep oom_messages_exporter_last_killed_
oom_messages_exporter_last_killed_pid{command="framework",container="nodejs",namespace="lovely-project",pod="awesome-backend"} 1.990532e+06
oom_messages_exporter_last_killed_pid{command="php-fpm: pool w",container="php-fpm",namespace="oldwebsite",pod="a-legacy-app"} 235362.0
oom_messages_exporter_last_killed_time{command="framework",container="nodejs",namespace="lovely-project",pod="awesome-backend"} 1.69484585513779e+09
oom_messages_exporter_last_killed_time{command="php-fpm: pool w",container="php-fpm",namespace="oldwebsite",pod="a-legacy-app"} 1.694845840949523e+09

Alerting rules

expr:

time() - max_over_time(oom_messages_exporter_last_killed_time{container!="None"}[Xh]) < X*60*60

where: Xh - a time period in hours for looking the last OOM Killer kill.

annotations.description:

the last kill was in ns={{ $labels.namespace }}, pod={{ $labels.pod }}, ct={{ $labels.container }}, proc={{ $labels.command }}, {{ $value | humanizeDuration }} ago

Production

# pip install oom_messages_exporter
# python3.9 oom_messages_exporter.py

Config

All runtime configuration provides by environment variables:

• oom_messages_exporter_PORT - an exporter port (default: 9001)
• oom_messages_exporter_POLL_INTERVAL - log polling interval (default: 1 (sec))
• oom_messages_exporter_MESSAGES_LOG - a path to log (default: /var/log/messages)

Dependencies

• python 3.9
• prometheus_client