The biggest thing I learned is give postgres a crap ton of ram especially if you use autosubscribe to communities bot. Nginx logs can tell you response times and timeouts, timeouts will be from slow queries and if you get a ton, need to bump up your RAM.

Use https://phiresky.github.io/lemmy-federation-state/site to troubleshoot federation issues.

Using the igpu might be problematic for transcoding if you need that. I’d recommend older intel / Asus NUCs if you want a mini PC. 3 year warranty, built for Enterprise, tall version has room for a 7mm tall sata SSD or HDD along with nvme m.2 SSD.

I think if you do Asus 12gen + they have another m.2 slot though it is the smaller one 2242. Doing all this you can upgrade it to 64GB RAM, 8TB m.2 2280, 8TB SATA SSSD, and 1TB M.2 2242. In homelab especially with mini PCs the limit is usually RAM / storage rather than CPU.

I got 4 11th gen with 64 GB RAM each and 32TB of SSD storage. I recommend avoiding QLC SSD as much as possible. Aim for TLC , MLC, or SLC. Higher storage capacity tends to be QLC or TLC, QLC has shortest endurance and slowest speeds.

Those disappearing messages aren’t just fun - they’re actually helping protect your privacy by not keeping your data around forever.

I highly doubt this is true.

I think something like this would do? You can search the list of supported devices there. Search by exposes power.

Forgejo supports a ton of repos including docker / OCI images.

Yeah works good until its under load which federation does have. Matrix and Lemmy both got like 20GB of RAM dedicated to the database on my servers.

There are postgres settings to reduce disk writes. There’s a max size and a timeout to write to disk. By default these values are on the lower end I believe.

I looked into the github repo for the open source version and found the images used. It looks like it uses the open source synapse app, so it suppprts whatever the synapse server supports. Unfortunately the operator wouldn’t fully support the new matrix 2.0 sliding sync / OIDC auth currently and since its the same synapse server it doesn’t support high availability which partially ruins the point of k8s.

That being said it does seem to have some support for workers, but it seems to be a WIP and that definitely could save some time setting up workers since it can get quite complex. They have a basis for making secific worker types where you simply specify the type and it would theoretically start the worker and have it properly work with your proxy.

I wonder if they will open up the core version to support it. Guess we will see.

Starter doesn’t support OIDC so free synapse I go even though I’d like a matrix server with high availability.

Servo is an active project still just not owned by Mozilla.

Yeah was kinda sad since android got like 3 releases in the last 3 days but SDK is not updated yet I guess. I’m hoping unified push will work with it better since it stopped working this month.

Edit also unread count / marking as read, that seems super broken in the older app.

Unfortunately the rust SDK / android version still doesn’t support native / simplified sliding sync. I updated synapse to v1.115.0 and cannot login. Apparently you have to use the proxy server sliding sync to login then toggle a developer setting, logout and log back in to use native one on android.

Its using OpenID so yes it would support those. But it would support a lot of others as well. Authelia, keycloak, and authentik are a few examples that are open source and self hostable. Its nice to have SSO for all your services even in a homelab.

NUCs are where to go. Intel chips good for transcoding and 3 year warranty. Had 1 die out of 3 die in 4 months and got a full replacement. Got another so I’m running 4 now and been about a year. Running tons of stuff and measured power to about $2.5/mo/pc.

Ah yeah it sucks to host on your actual computer, especially if you are using the laptop portably. I just wanted to point out storage isn’t too bad if setup right. Best of luck finding a good host for your needs!

Curious here how is storage an issue? Lemmy is 90% text and if you do the proxy for pictures its like 10%. If you use pictures from elsewhere and don’t use pictures it’s all text. I have had lemmy hosted for over a year now and it’s using under 35GB. I have a not that subscribes to top posts on larger instances so I should have a lot of communities loaded.

Edit: oh yeah Pictrs integrates with S3 api now so you can offload image storage to cloud for pennies.

At least one great thing about bitwarden, the passwords are stored on each device, so you kind of already have backups. That being said backups for vaultwarden is still beneficial.

Yeah it’s a fun ride.I run a cluster and the storage is replicated and distributed among them. 3/6 of my disks were QLC. One disk went bad super early and caused a bad latency but the other 5 disks kept performance up for the most part. Once I found out it was hardware I went with an enterprise SSD with a a 3 DWPD rating (drive wipe per day) and those beasts are so good. TLC m.2s seem to be okay too but I don’t think I’ll ever touch QLC again.

I have 24TB of wasted SSDs. I learned not to buy QLC ever again haha