Fwiw, @Potions plans to support multiple servers behind a load balancer. I’ll let them elaborate more on their overall plans if they’d like.
jam
dimitarvp
Any impressions on how well does their Durable Workflows + Workers (and their various connected services) work?
Just lately started pondering trying a personal project there. Free tier looks very generous.
0
derek-zhou
Maybe, right around the time when “SaaS MVPs with few customers” just became completely hopeless. 
2
dimitarvp
Does not CF allow you to f.ex. have a Dynamic Dispatcher, which is a Worker that allows you to route requests to other Workers via in-code routing and not just through URI matching? I.e. headers, secret values passed in there or form params or URL query params etc.? And then you put a Firewall / WAF protections in front of it? And that DDoS protection is included in all plans, free included?
Not sure how well their docs answer it though.
I have no super detailed view though, can somebody share some horror stories?
I am not going to defend the corporation which half the internet depends on. Of course they do have ulterior motives, but from what I’ve seen from them – they seem to be mostly good citizens. Open to be convinced otherwise.
Potentially, but I’d not care if I pay $10 a month vs. $13. (Where do you get a $5 VPS nowadays? Grandfathered plan? I have only found 1-2 EU VPS providers that allow somewhere between $3 to $6.50 and have seen a lot of negative reviews. Though I might have been unlucky.)
And I hate to use buzzwords but when I last researched the CF offerings, I see how can one “deconstruct” otherwise big and complex app into:
- Separate Cron triggers (like scheduled Oban jobs);
- HTTP endpoint Workers, stateless and stateful (Durable Objects);
- Databases: SQLite / Postgres / MySQL;
- Secrets / secret stores;
- Env vars;
- Rate limiters;
- Queue producers and consumers;
- S3-like bucket storage;
- K/V storage;
- Analytics database;
- Rate limiters.
These are the ones I checked a few days ago when I was researching CF for a personal project.
Don’t get me wrong. To me Elixir reigns supreme in monolithic apps.
But honestly? If I can neatly separate my app(s) and “deconstruct” them into the pieces above, I’d give it an honest shot.
The real problem is not hidden pricing IMO. The thing that would make me wake up in cold sweat is that if you squint even a little, you’ll have PTSD flashbacks of the nightmare that is the microservices “app” (more like, 20 separate GIT repos).
And even that is not 100% true. You can have Workers being able to call other Workers so not every separate thing will reinvent the same business logic. Though at this point you have to wonder whether you’re not introducing 5x friction compared to, you know, just have everything be in a monorepo and running under a bulletproof runtime like the BEAM VM.
That – the potential microservices hell – would be my argument against Cloudflare. Hence I intend to test-drive it for a personal project I had in mind for years, before offering this new architecture at work.
2
Potions
Forum Sponsor
Thank you for the mention! this was launched earlier this week: Clusters: run one Phoenix app across several servers - Potions
On fault tolerance, that changed this week. With Potions you can now run across several servers behind a load balancer, with distributed Erlang between the nodes for PubSub and Presence across the fleet (link above). For Postgres, Potions gives you a dedicated database server and backups, but not automatic failover. For now, you can point an app at a managed Postgres (Supabase, Neon, etc.) and run the cluster on top of that.
2
gtcode
If I were building a huge app I’d probably blend colo and/or dedicated for an elixir-centric system with cf.
Closest I got this year was an “presentable MVP” on the Elixir side. So, BEAM for the monolith all the way.
0
mudasobwa
Creator of Cure
I am running several MVPs on a single AWS instance. Deploy is git push to GitHub, actions would do the rest, including rollback if needed. All of them use the DB (postgres) running on the same instance.
One of these projects is a fully fledged social network, another one is Cure language site with a real playground, etc.
I pay for that kinda $40/mo which is already cheaper than CF. And yes, I have a plan B for the accidental highload.
Docker sucks, that is the virtue.
2
nature
For a project that is for a customer or business, I will tend to use CloudFlare as a lot has already been laid in place so long as the project requirements align but for personal projects where I am not constrained by time, I will drop CloudFlare & I think the extra deployment effort I will encounter is even a plus as this will force me to understanding deeper concepts of Elixir and the BEAM runtime
0
nature
With time, Elixir enthusiasts will come up with hosting solutions that will make deployment of Elixir apps simple
0
