I think there is a need for a managed postgres that is one step cheaper. For me as a solo developer it is quite pricey to enter at $38/month when I am not earning any money on my application yet. I love fly and would like to stay, but it is hard to ignore that digital ocean has a managed postgres for around $15/month. I wish you could introduce a small version for around $15/month. That way it is possible to grow into the more expensive solutions when money start coming in.
I’m using the free tier of Supabase at the moment. Their next tier is $25, which is also a bit chunky. So I don’t disagree with the sentiment, but there are many other options available.
Have you tried a DO database at the $15 tier you mention, with your Fly app? Why not use them for now?
That is an possible option to use something outside fly. Personally, I like to keep my whole backend (including database) in one “place”. Both for security and for performance. So if I move to Digital Ocean then I will move the application too.
But I appreciate your answer. Maybe I need to consider a hybrid solution (app in fly and database in DO).
Even though I’d like Fly to offer a cheaper plan, I don’t think it fits their strategy anymore. They’re likely focusing on bigger orgs, since that’s where the money is. For small projects or getting started, Neon and Prisma Postgres seem like the most affordable options right now.
We obviously have to respect fly and their business plan. That said, I think it is strange that a minimal Phoenix app with postgres now is >$38. That is probably going to completely stop developers from trying out fly.io. Atlreast, I would probably never have started using fly if that was the initial starting cost to try out the platform.
I’d doubt that, on the basis that Fly have gotten to where they are without a managed db offering at all.
The other thing to consider is whether SQLite and LiteFS might fit your need. It’s not a traditional database server of course, but if you’re running several hosts anyway, it synchronises all writes in real-time, so I’d expect it to be safe.
As it happens, I was going to try this for my app under development, and then I realised I had several experimental/learning things in flight already, and I didn’t need another
I’d doubt that, on the basis that Fly have gotten to where they are without a managed db offering at all.
I am not sure. When I started they offered the legacy postgres for $5 per month. And it worked very well. So for a normal phoenix application that was mostly off (min instance 0) the cost was $5 / month.
Well, a single host configuration has always been discouraged, and for good reason. So yes, it would work, but it was never safe. I’d say the minimum sensible cost is x3 that value for a cluster, and one has to factor in the time-cost of DBA tasks for node corruption situations.
Customers can of course still run single-node PG if they want to, for apps that are just in dev.
you can still run the unmanaged postgres for the same price, we’re not stopping you from doing that. for now the flyctl commands still exist, but in a future where they will be removed you can look at this repository that has the configuration GitHub - fly-apps/postgres-flex: Postgres HA setup using repmgr
or you can of course deploy postgres yourself without any of this; for a cheap (but not reliable) single-machine instance it should be quite simple.
Well, a single host configuration has always been discouraged, and for good reason. So yes, it would work, but it was never safe. I’d say the minimum sensible cost is x3 that value for a cluster, and one has to factor in the time-cost of DBA tasks for node corruption situations.
I agree. But I think there are two separate needs for databases in fly.
Testing out fly.io, Development, Small hobby project
Production application with real customers that are dependent on infrastructure
Fly.io makes most/all money on category 2. Whereas my point is that there is a need for 1. Both for attracting new users (to find out that fly.io is amazing) and for small development project that can grow and become paying projects in the future.
So if fly.io could introduce as cheaper tier of their managed databases, I think it would benefit both the users and fly in the long run. But that is just my 2 cents.
a big part of why we created managed postgres is because people kept putting their production data in cheap single-instance databases and yelling at us when the underlying host died and they lost data… or similar shoot-yourself-in-the-foot situations.
we do want to introduce a lower price tier for this use case, it just isn’t the priority right now.
$38/month is expensive for what you get (1GB and 25% of a CPU plus storage). You can run multiple databases in the cluster but with so little hardware it’s a hard sell.
Sure you get multi-region… but if that provided any uptime guarantees how come Fly SLA is only 99.9%?
