A collectd plugin which checks if given systemd services are one of
* state "running"
* state "reloading"
* state "state"
* state "dead" with a service type is "oneshot"
in each of these cases it sends sends graphite metrics of 1.0
.
Otherwise it will send a 0.0
value.
The plugin is particularly useful together with grafana's alerting.
Make sure Python dbus bindings are installed in your system:
- Debian/Ubuntu:
sudo apt-get install python-dbus
- Fedora/CentOS:
sudo yum install dbus-python
Copy collectd_systemd.py
to collectd Python plugin directory
(usually /usr/lib64/collectd/python/
or
/usr/lib/collectd/python/
). Add following snippet do
/etc/collectd.conf
:
LoadPlugin python <Plugin python> ModulePath "/usr/lib64/collectd/python" Import "collectd_systemd" <Module collectd_systemd> Service sshd nginx postgresql Service httpd </Module> </Plugin>
If your service has dash in the name, you need to wrap that name in double quotes:
<Module collectd_systemd> Service "celery-bots" "gunicorn-data" </Module>
Restart collectd daemon and open grafana web ui. Add a new graph with following query:
aliasSub(collectd.*.systemd-*.gauge-running, '.+systemd-(.+)\..+', '\1')
You should see all configured systemd services in the graph. Now it's
enough to add an alert for values lower than 1.0
to be paged when
services are down.
Following configuration options are supported:
Service
: one or more systemd services to monitor. Separate multiple services with spaces. Multiple services lines can be specified when they will be concatinated.ScanNeedReload
: monitor the needreload status of all units (off by default)NeedReloadIgnore
: Space separated list of one or more systemd units to ignore the NeedReload status of (empty by default) e.g.:NeedReloadIgnore "service1.service" "tmp.mount"
Interval
: check interval. It's ok to keep the default (60 seconds)Verbose
: enable verbose logging (off by default)
Each configured service (e.g. sshd) will be reported. If the value is less than one check:
systemctl status sshd
Each node will be report this value. If the value is less than one check:
systemctl status systemctl --state=failed
Each node will report this value. If less than one then identify stale unit by looking in collectd log file for an entry:
systemd plugin [info]: Unit needs reload: certmgr-renew.timer
or by running the command:
for U in $(systemctl --no-pager --no-legend | awk '{print $1}' ) ; do systemctl show -p NeedDaemonReload -- $U | grep -q yes && echo $U done
Install tox using pip or Linux package manager.
Type tox
to run tests.
On Redhat systems some selinux policy may be needed. Create a file collectd_systemd.te:
policy_module(collectd_systemd,0.1); require { type collectd_t; type initrc_exec_t; } dbus_session_client(system,collectd_t) init_status(collectd_t) init_dbus_chat(collectd_t) systemd_status_all_unit_files(collectd_t) allow collectd_t initrc_exec_t:service { status };
Create a file collectd_systemd.pp and install it:
make -f /usr/share/selinux/devel/Makefile collectd_systemd.pp semodule -i collectd_systemd.pp