Skip to main content

Is Railway Really a Conscientious Cloud Service? A Container Hosting Platform Review

Railway official website: https://railway.com/

1. Preface​

When you see that for just $5 per month, you can get an environment with 8 GB RAM / 8 vCPU for each service, isn't that exciting? Not to mention that according to official claims, this $5 also includes 100GB of storage.

This is what I saw after activating their service:

Hold on, things aren't as rosy as you might think.

2. Quick Database Deployment​

When you want to deploy a database service, it's really simple. Just right-click in the workspace to deploy:

And there are even options for Redis and MongoDB:

The first experience was indeed nice - just a few simple steps to complete deployment. The corresponding password is shown in Variables:

3. Database Deployment Region Limited to USA Only​

This is the case even after activating the Hobby Plan - only Pro plan and above can change this.

Connecting to a USA database from China has latency of at least 170ms+. Unless your server is also in Hong Kong, where latency is around 40ms.

This leads to the next issue - you'll want to create your own PostgreSQL service and specify the region, because you can choose Singapore machines when deploying with your own image.

Note

Later discovered that Hobby Plan can also change the service deployment region now, and migration is quite convenient.

4. Difficult to Custom Deploy a Desired Database Service​

Deploying a database with Docker is very simple. Using PostgreSQL as an example:

docker run -d \
  --name postgres \
  --restart=always \
  -e POSTGRES_USER=xxx \
  -e POSTGRES_PASSWORD=xxx \
  -v ~/postgresql/data:/var/lib/postgresql/data \
  -v ~/postgresql/config:/etc/postgresql \
  -p 5003:5432 \
  postgres:16.0

When you see the Docker image option on Railway, you might think you can deploy a fully customized PostgreSQL, configure some variables, and you're done. But it doesn't actually work that way.

In practice, the service can start, but I found two easily overlooked points:

  • You need to create a Volume and mount it to postgres
  • You need to specify the corresponding Volume in postgres
  • The configuration variables aren't as simple as just POSTGRES_USER and POSTGRES_PASSWORD needed for docker run

And you'll find you have no idea how to connect to this service. Even after configuring the port in settings:

But you still can't connect, even though the service is running, because the configuration variables also need these:

So even if you configure these variables, can you connect? That's unknown, because the following issue is even more frustrating.

5. Database Mount Volume Limited to 5GB Maximum​

This is really disappointing. The official documentation says there's 100GB of shared disk, but databases can only mount 5GB??

As shown in the bottom right corner, mount disk capacity cannot exceed 5GB:

It's like your boss promising you 5 months of year-end bonus, but when year-end comes, you only get one month. Guess what? The boss says: I'm saving the other 4 months for you.

After searching, I found others discovered this issue too: 5 GB or 100 GB disk on the HOBBY plan?

Their explanation:

Volumes are limited to 5GB, volumes are for persistent storage. The 100GB you're referring to is for ephemeral storage. Databases need to persist data between redeployments, so they need to use volumes.

This is clearly a threshold set deliberately to push users toward the Pro plan, which is really annoying.

Redis and MongoDB have the same limitation:

6. You Can Even Deploy SpringBoot Services​

Besides databases, Railway also provides many ready-made template services. As a Java programmer, I searched for SpringBoot and found it:

It actually works. Logs shown:

Of course, the request speed is... touching. You can test with this endpoint, valid until April 1, 2025: http://javaspringboot-develop.up.railway.app/

Other templates include various services, even Kali Linux:

7. 8H8G Might Burn Through Your Money Faster​

It's not that giving users 8H8G on the Hobby Plan is wrong - after all, users can set resource limits for each service.

But the Hobby Plan only has $5 monthly credit, which is a bit tight. Deploying a few more services easily exceeds this limit.

With only PostgreSQL and some test services deployed, my billing from March 10-15:

Estimated month-end cost is about $1.74.

So $5 can deploy about two and a half PostgreSQL instances. Without considering the cost limit, an 8H8G server can actually deploy many applications.

8. After Six Months of Deep Experience​

After six months of deep experience, Railway still feels inferior to buying a traditional VPS.

Here are the main reasons I strongly want to migrate away from Railway:

8.1 The Most Frustrating Point: Capacity Limits​

That 5GB limit is like the Sword of Damocles, always making you feel it'll fill up soon. And when you remember the official promise of 100GB, frustration wells up - you just feel like a sucker.

8.2 Network Traffic Costs Cannot Be Ignored​

Yesterday afternoon, there was massive database egress traffic - you can see it even reached 5GB.

So where did this egress traffic come from?

Just from this query SQL - a scheduled task running every 2 minutes, and that table only has 7,967 records. When I checked the traffic that night, I was incredulous - such little data, such high egress bandwidth??

This traffic directly generated over $1 in charges. Remember the Hobby Plan only has $5 monthly credit - if I hadn't noticed this traffic cost, by tomorrow morning half of that $5 might already be gone. And that month all my services were just 1 Redis and 1 PostgreSQL.

8.3 Latency Is Touching​

Even using the nearest Singapore node, connecting from a Hong Kong server to Railway Redis takes at least 70ms minimum - that's over 10x slower than local Redis deployment. If you deploy on Railway's US node, it reaches second-level latency.

What about PostgreSQL? Sorry, database speed is worse than wrapping it with Cloudflare Hyperdrive. Using Cloudflare Hyperdrive to provide a RESTful interface takes about 300ms+.

Direct connection from Hong Kong server to Singapore PostgreSQL node: 600ms ~ 1300ms latency.

8.4 Tasteless to Eat, Wasteful to Discard​

That $5 Hobby Plan credit really makes you hesitant. I'm always worried that deploying one more service will quickly exceed $5, even though I know they have detailed billing. But you know, that feeling of uncertainty makes you timid. After all, if costs take off, that heart-stopping feeling when the bill arrives is no joke.

Railway's cloud service feels like a beautiful, smooth-operating sports car with a tiny gas tank and expensive fuel - you don't dare step on the gas. The engine can hit 200 km/h, but you can only carefully drive under 50.

Functionally it "has everything it should" - deployment, logs, monitoring, variable management are all smooth, and the UI is cleaner than many veteran cloud services. But once you want to use it as your main production environment, you'll find latency, bandwidth, and billing models create an "invisible ceiling."

You'll hesitate: Should I spin up another instance here, or just move the task back locally? Should I use their database, or just buy a cheap VPS and set it up myself? This experience is like that dish on the table - smells good, tastes okay, but gets tiresome after too much, yet throwing it away feels wasteful. Ultimately, it's more like a "playground for testing new ideas" rather than a "reliable factory for long-term business hosting."