Feature parity with the node_exporter

[SuperQ]

It would be good for this exporter to reach feature parity with the node_exporter's systemd collector.

Once this issue is complete, we can remove systemd support from the node_exporter and link to this one.

[povilasv]

Thanks for creating this issue. This definitely needs to be looked at

[povilasv]

I've looked at it. Comparing v0.4.0 systemd vs prometheus/node_exporter@fa4edd7

We could potentially have feature parity in v0.5.0

I plan to add a README section on differences, so that people can easily migrate if they choose to do so.

From flags perspective:

Flags change from:

• --collector.systemd.enable-restarts-metrics => --collector.enable-restart-count

and

• --collector.systemd.private => --collector.private
• --collector.systemd.unit-blacklist => --collector.unit-blacklist
• --collector.systemd.unit-whitelist => collector.unit-whitelist

We don't have:

--collector.systemd.enable-start-time-metrics
--collector.systemd.enable-task-metrics

As these features are enabled by default, with no option to disable it.

From metrics perspective:

We have same exact metrics but different namespace (node_exporter has node_systemd_units, we have systemd_unit_info:

• unit_state
• unit_tasks_current
• service_restart_total
• timer_last_trigger_seconds
• socket_accepted_connections_total
• socket_current_connections
• socket_refused_connections_total

Differences:

• unit_start_time_seconds, we have additional type label
• unit_tasks_max, we have additional type label

We don't have:

• system_running - Whether the system is operational (see 'systemctl is-system-running').
• systemd_units, example:

node_systemd_units{state="activating"} 1
node_systemd_units{state="active"} 221
node_systemd_units{state="deactivating"} 0
node_systemd_units{state="failed"} 0
node_systemd_units{state="inactive"} 67

• node_systemd_version

Example:

node_systemd_version 244

We add:

• unit_info
• process_cpu_seconds_total
• unit_cpu_seconds_total
• process_open_fds
• process_max_fds
• process_virtual_memory_bytes
• process_virtual_memory_max_bytes
• process_resident_memory_bytes
• service_ip_ingress_bytes
• service_ip_egress_bytes
• service_ip_ingress_packets_total
• service_ip_egress_packets_total
• prommod metrics.

[SuperQ]

Awesome, thanks for the update.

Can node_systemd_units be derived from other metrics? Maybe it's unnecessary to add if you can do something like count by (state) (systemd_unit_state)?

For system_running, the typical pattern for an exporter like this would be to expose a system_up metric. We do this in other exporters like mysql_up.

[SuperQ]

One minor naming convention thing I noticed in this. We like to put the "unit" at the end of the metric name for consistency/clarity.

• unit_tasks_current -> unit_current_tasks
• unit_tasks_max -> unit_max_tasks

Kinda like process_max_fds

[sichvoge]

@povilasv Has there been any progress on this ticket?

[povilasv]

@sichvoge not really, busy with other stuff. Feel free to pick it up.

[sichvoge]

Thanks a lot @povilasv. Unfortunately, I am not an Engineer anymore and I watch this from a project point of view at the moment. :)

Anyways, thanks for all your hard work. Really appreciate it.