Does anyone run their own Lemmy instance on a pi? How was the process of setting it up? Were there any pitfalls? How is performance?
[Edit] So a lot of testing around. Compiling from scratch, etc, etc…
So far i have tried
- installing lemmy using rootless docker (on 0.17.3)
- compiling the image 0.18 docker image as arm
rootless docker did not work well for me. lots of systemd issues and i gave up after running into a lot of issues. I tried rootless docker for security reasons. minimal permissions, etc.
When trying to compile the latest lemmy image in arm, i ran into issues with muslrust not having an arm version. It might be worth investigating rewriting the docker file from 0.17.3 to work with 0.18.0 but i haven’t investigated that fully yet! I tried compiling the latest image because i wanted to be able to use the latest features
At the moment, I’m trying to set lemmy to run under bare metal. Im currently attempting to compile lemmy under arm. If that works, i’ll start setting up .service files to start up lemmy and pictrs.
Well, “ish”.
My experience with databases in general (granted, more the big ones than stuff like Postgres and mySQL) is that a lot if not most of the stuff that’s important for performance is held in memory (certainly they’ll tend to keep the most frequently fetched stuff in memory, along with the most used indexes) so I suspect the bigger Pi devices (with 4GB and 8GB) might just have enough memory to handle a good number of people doing common usage stuff (say, checking All in Active mode).
With a really big database and usage profile which has a random uniform distribution (i.e. any data piece is just as likely to need to be fetched as any other) then for the DB to be I/O bound in a Pi makes sense, but it’s my impression (or maybe its just me ;)) that Lemmy data access is very concentrated in a just a few things (which do change over time but the DB engine wll naturally adjust the memory cache contents for that kind of change)
From the little that I know about the structure of the Lemmy software, I expect it’s the image server that’ll have problems with slow I/O rather than the database.
Of course, all this is just conjecture, as while I worked in high performance computing, it wasn’t exactly done with Raspberry Pi devices ;)