Set up Metrics

[cahfofpai]

It would be very interesting to have metrics for the Transitous instance of Motis:

• how many API requests per day / per month / until now
• when are peeks in the amount of requests
• maximum number of parallel requests
• which service is the request for (routing, geocoding, ...)
• what is the usage of the system resources (cpu, ram, disk, bandwidth)

For example Grafana could be used for building a dashboard and Prometheus for collecting the data.

The request based matrics could be derived from the server logs:

• https://github.com/martin-helmich/prometheus-nginxlog-exporter
• https://github.com/nginx/nginx-prometheus-exporter
• https://medium.com/@mykprok/monitor-nginx-metrics-with-grafanadr-a-step-by-step-guide-6a3014c1783f
• https://github.com/nginx/nginx-prometheus-exporter/blob/main/grafana/README.md

For the system resources there are other exporters:

• https://github.com/prometheus/node_exporter

Maybe I could try to implement this myself.

What do you think about this?

[mlundblad]

I don't know if usear agent might also be useful to see the usage per app.

Not sure if other apps sets User-Agent to something specific (GNOME Maps sets "gnome-maps/" in the User-Agent header).

[cahfofpai]

I don't know if usear agent might also be useful to see the usage per app.

Not sure if other apps sets User-Agent to something specific (GNOME Maps sets "gnome-maps/" in the User-Agent header).

I also already had the idea for metrics per app, but left it for future improvements because it would need support by the clients. I thought about a separate URL parameter or similar with the app id, but the user agent would also be a possibility.

A topic to research for web clients would be whether they can set a different user agent for their requests, as normally it is set by the browser (also see https://stackoverflow.com/a/71172812).

[jbruechert]

I agree this would be cool to have. I think Spline already has a Garafana instance that we could probably use.
For the actual setup to collect the statistics I would welcome contributions as I personally don't want to increase the portion of my time going to sysadmin-style stuff in this project :)

(I can of course help when access to the server is necessary)

[traines-source]

Number of queries etc are intended to be directly included into MOTIS as Prometheus metrics, see here: motis-project/motis#696
Not sure if as soon as this is implemented, web server metrics are still that relevant.