cross-posted from: https://lemmy.cloudhub.social/post/347779
I am running a Kubernetes cluster for this domain, and I’m looking at more services to run (right now I have Mastodon and Lemmy).
I was considering WriteFreely and PixelFed, but they don’t seem to have an easy solution for running on Kubernetes (WriteFreely doesn’t even have a production-ready docker image).
Is anyone else running federated services in their lab? Do you run any of them on Kubernetes?
I run plenty of services in my lab on k3s. Most of them I had to manually build charts for and add them to my cluster. I doubt anyone has built charts for Lemmy, maybe mastodon. Anything that’s dockerized like Lemmy can be put into kubernetes, but it’s going to take some doing. Good luck out there, and of course if you get it working then I think the maintainers would be happy to get a helm chart merged into their repo.
Yeah, that’s the pain point - building and maintaining the charts.
Also, I know the charts likely wouldn’t have to be super complex, but I’m used to working with Bitnami’s charts that are massively complex - I just don’t have the time to go that in-depth.
I start simple with mine. Start with the deployment, just getting the pod up and running, completely stateless. Get that stabilized, once that’s happy then start moving onto the service, connecting to the UI. Then I finish anything else I can before adding in state - environment variables (first just in the deployment, then once that’s done worry about a config map and secrets). (Adding volumes in just complicates it because then you have to reset those volumes if you want to go back to a clean state).
As a list:
- Deployment (stateless) - get to not crash
- Service - connect to it
- Environment variables, extracting out what can be
- Secrets, adding those to the chart
- Volumes, adding in state
- Move to a nice clean setup with a clean values.yaml.
Finally, once your happy, then make it conform to other standards. Getting it stood up in k8s is the hard part, then customizing can come after.
That’s actually super helpful! I haven’t done much custom Helm chart-ing, and was kinda lost where to start. That really helps break the process down, and the tip about skipping state to start is very wise.
I’ve done… an annoying amount of them now. I hope my trials help you.
Be ready, it’s a very annoying and slow process, watching logs, figuring out why things are failing, debugging, githubs, everything. I just did one last week that was saying that it couldn’t write to /etc/passXXXXX, and it took 2 hours to track down that there was an optional command I could pass in that would change the running user of the container I was running (separate from the kube user). It’s a slog, but when you get it running it’s a rush of endorphins.
Biggest thing - kubernetes is a read-only file system compared to docker. So, for good devs they minimize writes to the filesystem unless they have to, and keep them localized. For bad devs, they write everywhere - and that gets sticky fast. So, if you’re getting write errors, know that there’s probably another volume you need to attach. The kicker is knowing which ones can be an
emptyDirscratch directory and which ones actually need to persist. If you have adocker-composefile it’s a great place to start, just set all the volumes toemptyDirto start off with.Good luck!
I think both of the ones I mentioned have
docker-composefiles, which I think I can convert withkompose convert? I guess from there I would follow your steps and then start parameterizing it once it’s running properly.Thanks! I think I’ll start trying out PixelFed tomorrow.
This all makes sense to me since we deal with it at work. I would maybe add a service vs route point to differentiate things like UI that need external exposure. The main difference is we use kustomize instead of helm. Out of curiosity if you had any experience with both and why did you settle in helm?
I don’t like helm, so I use nix to maintain my fediverse deployments in kubernetes. Typically that’d just autoupdate itself to new releases, but for lemmy specifically I upgrade by hand nowadays since one release some time ago broke my deployment and its schema change was incompatible with the automated rollback.
My setup is a combination of https://github.com/farcaller/nixdockertag (auto-updated docker imagesfor things where I fully own the deployments) and https://github.com/farcaller/nixhelm (for helm charts that I either consume verbatim PR have local patches on). Both just auto update nightly thanks to github.
If they’re containerized they’ll run in a Kubernetes cluster, I’m sure you know this. Why not build the K8s infrastructure you need to get them functional? Or maybe Helm charts if you’re up for it.
Oh, I know I could get them to run with enough work. I just don’t have that much time to spend on initial implementation and upkeep of the charts.
I’m using FluxCD, which I believe can do deployments of plain Kubernetes manifests, but that still requires a decent amount of overhead to keep up to date.
I just don’t have that much time to spend on initial implementation and upkeep
Well k8s is a poor choice of platform for you :D
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
Fewer Letters More Letters CSAM Child Sexual Abuse Material HA Home Assistant automation software ~ High Availability HTTP Hypertext Transfer Protocol, the Web k8s Kubernetes container management package nginx Popular HTTP server
4 acronyms in this thread; the most compressed thread commented on today has 5 acronyms.
[Thread #740 for this sub, first seen 6th May 2024, 19:45] [FAQ] [Full list] [Contact] [Source code]
Using different federation protocol, but matrix wservers ould be the other big one.
Edit you also mentioned trouble creating them. I suggest looking into operator hub and using operators for postgres and redis and auth (keycloak?). This can get you down in the rabbit hole for making everything highly available too.
Yeah, I used to host a Matrix instance - could do that for this one too.
The issue is more about setting up the Kubernetes manifests and templating them. I usually use the chart’s built-in postgres and redis config, though using an operator would make it more scalable for sure.
I’m using Authentik for auth, but I do also like Keycloak.
Yeah it’s a bit of work sometimes. Synapse matrix kinda sucks too their philosophy of no environment variables for secrets. I ended up making an init container that hijacks my config map and I jet’s the environment variables into the config
They store the secrets in a file? Gross. What a poor way of handling that. Pretty sure environment variables would be more secure. Especially in Kubernetes.
Yeah I want to switch when other implementations catch up. Unfortunately I think that will be some more time especially since you can’t migrate from synapse and have to start from fresh. One day though!
I did the same for Lemmy at one point then found out all the configs are mapped to environment variables my convention. My Lemmy setup is the most advanced, but it has HA postgres, and all of its modules separated and HA. The proxy setup for it in k8s was rough but I eventually got it working directly on ingress-nginx too.
Huh, do you have your lemmy config documented somewhere? I keep running into issues with it and I’m not sure which component exactly is failing, but it’s annoying. I’m using this helm chart currently: ananace/lemmy It works, but I don’t have pict-rs setup in HA either.
I got all my yaml files source controlled privately right now but I can share if you want them. I disabled Pictrs around the time of CSAM attacks and have yet to bother enabling it again haha
I disabled Pictrs around the time of CSAM attacks and have yet to bother enabling it again
Uhh… what?? When did that happen? I thought pictrs was a requirement also…
It’s not kuberneties, but I run a family sized yunohost. It’s great at installing and updating webapps. They have an awesome selection of federation apps like mastodon, writefreely, misskey, bookwyrm, and more.
For less than 5 users, I personally don’t need kub, but it f I were to scale I would probably go that direction.
I’ve seen that around, but I prefer to run my own services instead of relying on a ready-built system like that. I find they don’t offer that much customization options usually.
Found a couple, you should be able to find more on artifacthub.io
