Post by @cora@hachyderm.io

Question for the #homelab nerds out there.

I have a running family of projects (well more like a breakfast club of misfits) that I've been deploying to my "lab" (by which I mean, currently, a mediocre #raspberrypi), currently stitched together with docker-compose.

An important underlying theme to this effort, and a reason why I'm sticking with the pi, is a focus upon resource-constrained, high-efficiency, self-contained, open-source computing. Kinda the opposite of the vibe of the times, I know, but I think it's important.

I suspect I'm far from alone in this, and I'm curious: what are other folks working on projects like this use for observability? So far, I'm cobbling my own together with shell scripts and instrumentation in the code (using #tokio tracing if it matters).

I don't really want to go down the "standard" bobsled run that ends with an army of proxmox vms and various free-tier accounts on cloud-based services... again, resource-constrained and self-contained. It feels like we used to have tools like this, when every server was a pet; surely those haven't gone away?

Search results, unfortunately, are twenty pages of "The N best homelab observability stacks in 2026! (Number N-M will suprise you)" AI-generated marketing blogslop, all of which I've sampled launch an aforementioned sled.

So, asking here. Tips solicited. Retoots welcome.

#askfedi

#raspberrypi #tokio #homelab #askfedi

15 Replies

1 Boost

Elton Cora (Honorary)

@cora@hachyderm.io 5 hr. ago

Hey thanks everyone for reaching out with recommendations! I decided to start with a Victorialogs eval, and in a couple of hours that had turned into a full deployment including Victorialogs, fluentd, node_exporter, and postgres_exporter, all wired up and chatting and presenting data.

I'm surprised how quickly this came together. Still need to add metrics endpoints to my homegrown images, and don't have alerting set up yet, but aside from that it's humming along.

I really appreciate the suggestions. Few minor nits to work out, and I still intend to play at Beszel and Dockhand as well, but this feels like an accomplishment.

#homelab #raspberrypi #tokio #askfedi

MyonlinePi

@admin@myonlinepi.uk 8 hr. ago

@cora I mainly now just rely on Bezsel for monitoring together with Dockhand for helping to manage the Docker environment. Both are lightweight and easy to deploy with Docker compose. https://github.com/henrygd/beszel

GitHub

GitHub - henrygd/beszel: Lightweight server monitoring with historical data, docker stats, and alerts.

Lightweight server monitoring with historical data, docker stats, and alerts. - henrygd/beszel

happyborg

@happyborg@fosstodon.org 10 hr. ago

@cora do you see Proxmox as implying third party services? I avoid those too, and am resource conscious, and also want reliability and easy maintenance (eg fail over and backup), so I run Proxmox on Dell OptiPlex mini PCs.

Elton Cora (Honorary)

@cora@hachyderm.io 2 hr. ago

@happyborg I don’t see it as implying third party per se; it’s just that everything I saw recommending it pointed down a path of increased complexity in support of massive scale, and much of that did include integrating third party bits.

Also didn’t mean to poo-poo it, because it does look cool. I need to learn more about it of course; like I said, I’m trying to find ways to do as much as possible with as little as possible. If it can support that effort, could be cool.

happyborg

@happyborg@fosstodon.org 1 hr. ago

@cora that's cool. I just wanted to clarify. You may prefer more minimal and that's good too.

Snuggle2476

@Snuggle2476@mas.to 10 hr. ago

@cora Another vote for Prometheus/Grafana here, but on your footprint question: the collection side is featherweight, node_exporter and textfile collectors are tiny Go binaries. The RAM cost is really Prometheus's own storage, which is why the Victoria* swap upthread is smart. The piece nobody's mentioned yet is alert delivery. I run Alertmanager to ntfy, tiny and self-hosted, and it pushes straight to my phone. Mine's on a real box not a Pi, but ntfy runs fine on one.

Elton Cora (Honorary)

@cora@hachyderm.io 9 hr. ago

@Snuggle2476 Good to know. I hadn't given a _ton_ of thought to alerting yet, but I've used alertmanager in the past and it is fine.

Michael

@mmeier@social.mei-home.net 10 hr. ago

@cora I've been running Prometheus+Grafana for quite a while. Not the most resource-effiicent deployment, but versatile.

And for logs I'm running FluentD + Loki + Grafana.

Elton Cora (Honorary)

@cora@hachyderm.io 10 hr. ago

@mmeier Thanks! What kind of hardware are you throwing at it? I'm not religiously opposed to those two, and if the deployment and footprint can be stripped and tightened well, the would be candidates.

Michael

@mmeier@social.mei-home.net 10 hr. ago

@cora Right now, a Kubernetes cluster with 11 RaPis, 3x Pi 5 and 8x CM4 8GB - so not exactly the most resource efficient thing. 😉

But I previously ran that stack on a low power x86 machine together with other Stuff like Gitlab and Nextcloud.

Prometheus polls 21 machines for host metrics as well as the k8s cluster and a couple of apps. And it uses 0.125 of a Pi 4 CPU core on average, with a few spikes when I open Grafana and actually look at metrics.

1/x

Michael

@mmeier@social.mei-home.net 10 hr. ago

@cora It uses around 780 MB of RAM on average, but that should correlate pretty well to the number of metrics you're gathering.

Grafana on average uses 0.08 of a Pi 4 core and uses about 180 MB of RAM, again depending a bit on usage.

Loki is at around 0.08 of a Pi 4 core and around 200 - 300 MB. But again, that's probably going to be lower when you've got only one machine.

Hm, and FluentD obviously has a memory leak right now, because that's at 790 MB. 😅

2/2

1 more reply

Pete Keen

@zrail@hachyderm.io 11 hr. ago

@cora I use Gatus for basic health checking and I run VictoriaLogs and VictoriaMetrics, both in standalone mode. No Grafana, the built in reporting has been enough for me.

Elton Cora (Honorary)

@cora@hachyderm.io 10 hr. ago

@zrail Thanks... I did come across Victoria* on one site, and I'd forgotten about it, but IIRC it did seem to meet high-level requirements. Will take a look.

Paul

@plwt@mstdn.social 11 hr. ago

@cora My setup sounds very similar to yours, so first off - you are very much not alone.

Further to a recommendation made here on the fediverse, I am using https://dockhand.pro/. Easy to setup, works really well for what I need.

View

Elton Cora (Honorary)

@cora@hachyderm.io 11 hr. ago

@plwt Oh thanks, that seems comprehensive... will take a look. Thanks!