Ask HN: Small teams and solopreneurs, how are you hosting your apps? | Hacker News Reader

Ask HN: Small teams and solopreneurs, how are you hosting your apps?

29 pointsaosaigh51 comments

I'm a solo dev working on both my own apps and a product as part of a small team. I'm interested to hear how others are running their apps?

- Self hosted on a VPS? - Managed infrastructure like AWS, Fly, Render, Digital Ocean App Platform, GCP? - Full-stack platform like Vercel, Firebase, Supabase?

View on HN

Buy some used servers on eBay, at least 3 will do for redundancy, then get 2 internet connections. Set up another server to act as your opnsense firewall/router and bond the two internet connections together. Make sure to buy a pair of UPSs for power back up. Install kubernetes and rook ceph with about 12 hard drives (redundancy) then find a friend who can host another server for offside backups. Dont forget the tape drive system for your second media. Start deploying gittea so you can host your code. Use argocd so you can have a build system. Deploy dapr so you can host all of your microservices in a standard way. Deploy otel and alertmanager so you can hook it into the only cloud service you need: OpsGenie (don't forget to winge the whole time you use it)

That way you can write an article about how much cheaper self hosting is compared to the cloud and get your next round of investment.

On a serious note. The next startup I do I'm going to just use supabase.

Only about 10% of this comes off as sarcastic in the context of HN.

Just one server and skip the redundancy. Then it's kinda manageable

Self-hosting using FTTH is many orders of magnitude more stable than home based internet used to be.

A lot of the description above can be run pretty easily with Proxmox, which itself is a web based gui to setup said redundancy (even can run it as hybrid cloud with failover to the major cloud providers).

This approach can greatly simplify and reduce dependency formation on one cloud service in particular.

Fully managed so I can spend my time on actually building features. In my case, AWS is my go to cloud, and even with a couple of thousand users, Lambda for compute, DynamoDB for database and SNS+SQS for eventing is costing me less than 5€ per month. Yes, there are risks with serverless if you get DDOSed or whatever, but it’s a risk I’m fine with, and can mitigate with gateways in front if necessary. And Lambdas are not locking me in to AWS since I’m running “full” ASP.Net apps in them, so hosting them on actual compute platforms is an easy switch.

Very similar tech stack here, with all the same motivations. Biggest differences are I'm using RDS for the database and have a slightly different approach for serverless compute. I started out with Lambda, but the cold start times were bugging me so I moved production hosting to App Runner, which is the next best thing (eliminates cold start without breaking the wallet). Kept staging environment on Lambda. Also using the "fat lambda" approach with ASP.NET apps, so no lock-in here. Spending about $35/mo total - RDS is the biggest contributor, then App Runner and other small things.

I also have an SSR app for the homepage which required some dynamic functionality. Basic Node app hosted in a Cloudflare worker essentially for free, hitting the App Runner hosted API for data.

Interesting. Did you migrate to a serverless architecture or was your app serverless from the start?

It was serverless from the start, although there isn't a whole lot of serverless unique code in it, apart from some event handler bootstrapping to handle SQS triggered Lambdas. Everything else is just standard .Net that would look the same no matter where it's hosted. Huge fan of separating infrastructure concerns from the rest of the app to not lock yourself in so much.

Snap, I'm using CDK to set it all up too which makes everything pretty easy to manage.

Rancher k8s + Helm charts for HA services (Redis Sentinel + Patroni Postgres). Nightly snapshot backups + streaming Postgres WAL segments to S3.

I’ve never understood the k8s hate. We have been running this for a few years and it is rock solid. We can bring the entire cluster up on any provider anywhere in the world in about an hour. The DR is great.

Self-hosted on a VPS is the first option for me, it's cheap and you have total control on everything. It's valid in my case because I know my way through infrastructure and configuration, so a panel isn't too much gain.

You sometimes can even get those for free - https://free-for.dev/#/?id=major-cloud-providers

For example, I have some (really) free instances on Oracle Cloud that I host pages, experiments, and so many things.

S3 buckets for the SPA's, AWS Lambda, API Gateway and EC2 for a MySQL instance. Some SQS to help with orchestration of the operational stuff. Have about 5k MAU and total cost barely breaks 40$ a month.

Chose this because it's the stack I'm super familiar with.

Hetzner + docker swarm can take you long way for a fraction of cost of any next alternative

And for a fraction of a mental burden and stress of the lightest k8s setup you can imagine

I did something like this for a while when I was messing around with my personal website, but with one node in Hetzner and other in Linode. It was a really nice way of managing services without bringing a full kitchen sink factory factory.

The only thing I found lacking was an easy way to share volumes across nodes. There's plugins and stuff, but for me it was not worth the effort.

I ended up just ditching Linode and having everything in a single Hetzner VPS. I can do this because it was mostly experimenting and didn't really need to have more than one server. But for this specific case, the volumes thing was what ended up tipping the balance in favor of only having a single server because it wasn't worth the effort.

How often solopreneurs can’t get away with just 1 beefy machine? Or at least with just 1 vps provider to share extra volumes

I run 2 rails websites with hundreds of real daily users on a single shared 2cpu/2gb vps, inlcuding 1 staging env, postgres db, memcache for caching.

We have our own servers (mostly consumer pc parts) and run lxd to slice them up into vms/containers. Deploy to those using kamal and expose the web server port through cloudflare tunnels. With simple scripts to periodically compress and upload the storage volumes and database dumps to Google drive, most of our less critical apps cost about $0 to keep running. For the higher criticality apps, we also deploy them to Linux vms using kamal, but use a commercial service (linode).

Spoiler: Shameless plug, but might be helpful for people

I understand most of the commentators, as well as OP, are probably looking to not spend too much time and money on their hosting infra. While that makes sense in the beginning, there might be a point in your successful journey where you want to hand over your operation duties to someone skilled and focus on building software because your business starts depending on safe and well-maintained infrastructure.

If you're looking for something between Heroku and AWS (both in terms of pricing and scalability) but based on K8s, with direct access to skilled platform engineers and personal support, you might want to check out https://www.ayedo.de/cloud/

Dokku on a Hetzner instance, can handle 10s of apps if I need. Very simple to setup and manage. Deploys with git push.

For static web apps like my portfolio and blog[1], I use GitHub+Netlify.

I try to use AWS Lambdas as serverless backend as much as possible, for API products like [2].

If a server is absolutely required like running a Ghost blog/newsletter [3], I use AWS Lightsail mainly because I'm familiar with AWS in general and it offers generous free tier.

[1]: https://ivylee.github.io/

[2]: https://www.ycverify.com/

[3]: https://www.signalstalk.com/

Render has been a great heroku replacement. I thought I would miss the “add-ons marketplace” but it’s actually been fine. Costs are reasonable for our startup and they seemed to have fixed their odd Cloudflare downtimes of the past.

Docker on a DigitalOcean VPS. I've never been a fan of managed stuff, but then again I'm mostly coding for fun, not because it's the best use of my time.

This is not for a business, but I am working on a side project that uses node.js + sqlite and the entire stack will run on a single Lightsail instance ($5 per month)

I use Coolify on Hetzner bare metal.

I still like individual VPSs, as I have more control over the cost, monitoring, backups, etc.

So, I've found that a mix of VPS / Cloud is great for me.

I have VPSs with Ionos, they're cheap and for the rest its all over GCP, functions, scheduler, firestore, storage, CDN and Cloud run are convenient tools.

Namecheap for domains, feels reliable.

Maybe switching to Hetzner for VPS if I need bigger ones, feels reliable and good priced.

AWS for the important stuff: DB, core services

Cloudflare/Vercel for the UI

Increasingly moving the important stuff to Cloudflare

My setup is a digital ocean box with docker compose scripts and watchtower for deploys. All I have to do is push a new image and it will deploy in prod. Database is as aged Postgres cause that’s all that is really important. Cloudflare for the smaller services.

Currently using heroku for a backend server for an app I'm working on, in the past I've played with a lot of different options, but most of them are too complicated and expensive if you don't actually have paying users.

Netlify for the front-end, google cloud for extended storage and Akamai for hosting. Akamai is great as they are less customizable than AWS (which is a good thing if you don’t need extra config) and much less expensive.

Extremely happy with supabase here, it has mostly eliminated the need for a backend, letting me just focus on making valuable stuff for my users. I also ended up making my entire app realtime just because it was easier

Cloud offerings are incredible for small teams. Don't even bother trying to do your own thing until you can clearly articulate the savings and factor in the risks and costs.

Odds are you'll never need to move.

Hosted a few small but profitable web sites on Hostgator since 2005 - cheap and no need to worry about infrastructure beyond hosting files. Only problems have been wordpress sites getting hacked.

I try to stick to serverless rather than trying to manage any infrastructure

1) Google App Engine Standard. Datastore for DB.

2) For anything that won’t work with the above, then it’s Google Cloud Run

Render. Happy with it for now as a prototyping / MVP tool. Won’t be able to continue forever due to HIPAA requirements and not sure how it scales anyway.

Combination of labs, colo, and partnerships with other parties.

Varying types of platforms and degrees of control. Ansible is the most unanimous thing I can name

I host some very small things on Firebase but I'm mostly using Render for projects these days (I switched from Heroku)

Fly.io.

One command and it’s in another region, with my data thanks to litefs. It works great for my read heavy e-commerce app

Is there a better option than heroku for under 10$ a month for python app and db support ?

Render is great but is more like $20 a month for app and managed Postgres

I'm using Modal (modal.com). Works with my Python heavy stack.

1GB RAM DigitalOcean VPS

Vercel after moving away from Docker containers running on AWS.

What are the advantages of Vercel over Cloudflare pages?

Netlify for front-end and codehooks.io for the backend.

ELB with WAF -> Fargate -> RDS

VPS + Cloudflare CDN, R2 and Workers

Steam! No server at all save that.

Cloudflare Workers and Fly.io

Fly.io + EdgeDB cloud.

“rails s” on a droplet

im considering Railway due to not requiring credit card, but have not decided yet so cannot tell if its good or not

Reply on news.ycombinator.com