A javascriptless check was released recently I just read about it. Uses some refresh HTML tag and a delay. Its not default though since its new.

As long as its not configured improperly. When forgejo devs added it it broke downloading images with Kubernetes for a moment. Basically would need to make sure user agent header for federation is allowed.

Yeah operators are extremely nice. I used bitnami images before operator in HA setup and it would fail all the time. Using operators I have like 12 postgres DBs and 0 issues for well over a year.

Btw, self hosting communities tend to shy away from k8s due to complexity and generally enterprises use it so less selfhosters use it. I wanted redundancy and to keep me learning tech, less crazy people tend to go with proxmox for clustering in these communities.

Some operators for postgres is free for non commercial use, so its not truly open source even though its source available, one is crunchydata.

pgvecto.rs v0.4.0 I assumed was just not renamed. They have one for vectorchord too

Operators are good ways to support applications in CLI that aren’t easy to setup in a cluster by default. You can make these databases redundant by setting replica higher than 1 and applying it, operator copies the db data and makes a new replica. It also helps with backups and restoring too.

If you are using Kubernetes, I highly recommend investing time into installing an operator. The best open source one with less restrictive licensing is cloudnative pg. VectorChord builds official images for CNPG that includes the extension.

I didn’t think of using read only replicas, that would probably be a very good way to go since its probably 80%+ of actions are reads. Thanks for answering, I am excited to see the how lemmy grows and thanks for all the devs hard work!

I 100% agree with this and there have been great strides since I started using Lemmy ~v0.17! That said at some point optimization will have lower returns and have a higher effort to put into and once a community grows extensively it likely might not be enough, so I was curious to what you guys were thinking at that point, something like Ctius for sharding postgres?

It is a k8s cluster and using ceph for all of my storage so the latency from that I bet is the largest reason and upping the memory offsets the disk writes. i also have another postgres DB syncing as a fallback for high availability. Fortunately after tuning the database and giving it enough RAM my instance has been running pretty stable for over a year without any changes.

I am also using less powerful computers for the entire infrastructure (not server grade) which brings to the point of having horizontal scaling on database I imagine will be a growing need with growing instances, communities, and users since it can be cheaper to run multiple smaller spec servers rather than a single with the added benefit of high availability.

Yeah I used pgtune as a base and found more memory needed to be assigned to certain spots especially to keep federation with bigger instances, otherwise timeouts would occur resulting in my instance being constantly behind.

That said I read postgres 17 is much more memory efficient, though I have yet to move my lemmy database to it yet since its the largest haha.

On the server perspective, I have a question, what are your thoughts for horizontal scaling on the database? This seems to be the biggest limitation and requiring higher spec hardware to scale especially for the bigger instances.

My tiny instance for example I give over 20GB of RAM just to postgres to make it perform efficient enough.

There is an issue open requesting this… I been following it for a while.

Issue #2631