Comment on Observability

• curbstickle@anarchist.nexus ⁨3⁩ ⁨weeks⁩ ago

  Prometheus and Grafana are my typical recommendation there, with the podman exporter. Prometheus for time series data, Grafana to visualize and alert.

  One thing I think many folks get wrong is thinking about Grafana (or any other dashboard) as the way to watch for problems, but that requires you to look at them. They should be thought of as a way to evaluate for conditions that there was a problem, then create an alert based on the indicators. This way you can address it before it becomes a problem, or at a minimum be aware that conditions are leading toward a problematic state.

  If its repeatable and managed with a container restart or something, you can automate that from the alert, and never have to look again.

  If its something you can’t control (for example, availability of a 3rd party service), you can use it as a way to be notified of the status change. I have a custom exporter I wrote for work to keep track of some services that result in a bunch of posts to Teams by people asking if its working for them. So I have an automation that posts to Teams when a service has a problem so my IT guy doesn’t need to respond to all of them individually.

  So I’d recommend focusing on good alerts and automations over beautiful dashboards first (though dashboards can be a lot of fun to make!)

  original

  • captcha_incorrect@lemmy.world ⁨1⁩ ⁨week⁩ ago

    Hmm, I wonder if I even need a dashboard (other than for fun). Perhaps I should do as you say and first look at what I really need to make sure is working (backups) and then monitor that and set up alerts for it to begin with. Thanks for the insight!

    original
  • farcaller@fstab.sh ⁨3⁩ ⁨weeks⁩ ago

    I’d suggest VictoriaMetrics over pure Prometheus - better ram footprint is pretty useful nowadays.

    original

    • curbstickle@anarchist.nexus ⁨3⁩ ⁨weeks⁩ ago

      They played a good bit fast and loose with their claims some time back (even pulling a few things after an “oopsie - decimal point in the wrong spot!” type issue), and I kind of wrote them off after that. I’ll have to check in on them again.

      original

      • farcaller@fstab.sh ⁨3⁩ ⁨weeks⁩ ago

        I’ve used VM in my homelab for a while - it’s genuinely a smaller ram footprint, especially on the ingress path. You can even scale query separately if required.

        original
• surewhynotlem@lemmy.world ⁨3⁩ ⁨weeks⁩ ago

  They are useful for diagnosing recurring crashes.

  original