god, OVH was fucking terrible when I tried them a few years ago. a lot of fediverse admins swear by them now though so either something changed or the bar for usable hosting and global outages lowered
and reading this back, it feels like there's so much free space on that CX21 maybe I should do a best-effort WriteFreely on there, just to justify the budget for the node
the actual Lemmy service running awful.systems isn't clustered yet -- all of Lemmy, its Postgres database, and pict-rs run on a single Hetzner Cloud CPX31 with quite a bit of CPU and memory to spare. some services (mostly static hosting for a couple of things, plus the staging environment for upgrades and configuration changes) are offloaded to a CX21 that's definitely overprovisioned for the usage it gets. the CPX31 hosting us is behind a LB11 load balancer for future expansion -- if I need to stand up another instance of the Lemmy frontend, I can live reconfigure the LB11 to round robin onto that host without any downtime. there are downsides to using Hetzner's LBs -- they're extremely inflexible and basically can't be configured outside of a typical "round robin and terminate TLS" use case, though they're very nicely automated (they'll even manage Let's Encrypt for you) for that use case.
as with all federated services, a Lemmy server that's being used in any capacity will slowly fill its available disk space with posts and associated data. currently we're still rolling with the storage included with the CPX31, but Postgres and especially the rather inefficient image cache are gradually filling that disk. part of the plan for the deployment is to either offload the image cache to object storage (which can be extremely cheap, but definitely do the math on egress charges) or, more likely (because it helps keep us portable between cloud vendors), I'll expand the LVM for the node's disk onto a Hetzner volume when we get to around 75% capacity.
if you're looking at establishing an instance along these lines, make sure you look at rate limiting first if you run into any performance issues. before considering any upgrades, check your access logs to make sure you're not seeing a spike due to malicious traffic. ActivityPub is a unique challenge to rate limit properly since some of your endpoints will always have a ton of repeated, automated traffic from other instances, but there are a few guides out there that have good defaults, and getting this right before you have a ton of users will save you time later.
other than the above, we have an external email service and backups that I can provide more detail on if you need recommendations as you get closer to rolling out your deployment.
obligatory: I don't recommend Hetzner as a company, but David and I have yet to find a host with comparable pricing that isn't somebody's hobby or Oracle, a company neither of us will deal with due to personal experience and industry reputation. the above runs about $25/month at the prices I get for Hetzner's resources (I think on Cloud they lock you into whatever rate you were at when you joined, so mine are cheaper than the ones on their main site), but you may get better value from a server auction or other host depending on your needs.
this is what the cloud and its enshittification has taken from us
shared root on ad-hoc hardware doing fuck knows what (but it’s probably lemonparty)
and here’s hoping the American nationalist devs contributing on behalf of their military-industrial complex employer (hello Anduril) take a hint from this and also fuck off to their own communities where they can bully each other for no fucking reason
they won’t because the cruelty is the point for fascists regardless of nation, but here’s hoping
oh no, a bunch of nationalist pricks might stop fucking up our community spaces. I might never have a proud Russian gatekeep my contributions ever again! no please don’t go
oof, I’m sorry. it’s so hard to get capitalists to understand the nature of what they’re enabling, especially if it seems to be working in the short term. it’s the most frustrating thing during a bubble — it taints every decision the executive class makes, and enables grifters to get away with obvious shit even over objections from people who know better.
hah of course!
I don't think you want to hear my opinions on what the left wing thinks is obvious :-)
Also, I am neither left nor right wing, as I'm a libertarian. I believe in the principles in the Declaration of Independence, the Bill of Rights, and the system of checks and balances set up by the Constitution.
it’s just really surprising to see the political takes of a 13 year old come out of the 65 year old who created the least successful C variant
lemmy explores new frontiers of ux design
the specific frontiers being explored are the parts of Oregon Trail that kill you
cool, time for this account to disappear too
a quick interest check: I kind of want to use our deployment’s spare capacity to host an invite-only WriteFreely instance where our regulars can host longer form articles
…but WriteFreely’s UI is so sub-optimal the official instance (write.as) runs a proprietary fork with a lot of the jank removed, and I don’t really consider WF to be production ready out of the box.
we can point the WF backend at arbitrary directories for its templates, page definitions, and static assets though, so maybe I could host those on codeberg and do a CI job that’d pull
mainevery time it updates so we could collaboratively improve WF’s frontend? it’s not a job I want to take on alone (our main instance needs to take priority), but a community-run WF instance would be pretty uniquethe pros of doing this are that WriteFreely at least seems to have very slim resource requirements and it’ll at least reliably host long form Markdown on the web
the downsides are again, it’s janky as fuck (it only supports Mailgun of all things for email, but if you disable that the frontend will still claim it can send password reset emails… but it’ll check the config and display an error if you click the reset link??? but they could have just hidden the reset UI entirely with the same logic???? also I don’t like the editing experience), and it’s not really what I’d consider federated — it shoots an Article into ActivityPub whenever you post, but it’s one-way so replies, boosts, and favorites won’t show up from ActivityPub which makes it feel a bit pointless. there might be a frontend-only way to link a blog post to the Mastodon or Lemmy thread it’s associated with on another instance though, which would allow for a type of comment system? but I haven’t looked much into it. write.as just has a separate proprietary service for comments that nobody else can use.
this definitely won’t replace Wordpress but does it sound like an interesting project to take on?