Ask HN: How do open source apps like ghost blog run their multi tenant version? | Hacker News Reader

Ask HN: How do open source apps like ghost blog run their multi tenant version?

4 pointsnstart6 comments

This one might have a well known answer but I'm not entirely sure how to find it. I've searched quite a bit but nothing gives quite the right answer.

Basically ghost blog comes with a large set of features in its open source version. They make this version available on their own paid hosted option as well. But their source code doesn't point to a hosted multi site version. I was wondering how this is done architecturally.

On a slightly similar note, how do they maintain the open source version with basic features with a closed source version with premium features?

This is done with a number of open source projects. Was wondering how it's done architecturally.

View on HN

There are a lot of ways to do this.

Most web applications have a parameter in them that points to one or more database servers of some kind, so you can create a new instance by doing, say, "CREATE DATABASE" in mysql and installing the software in a directory on a web server configured to point to the database. If there is a need for scratch directories and other resources that can be handled pretty much the same way as the database.

The provisioning system then is completely separate from the main body of the app so it is easy to keep one private and open source the other.

With the above you can often host a huge number of instances on one machine without any exciting tech (I was doing this at least as early as 2003) but obviously the idea could be implemented with containers, cloud servers, etc. today, for instance you create a cloud server for the user and auto-provision the software to the server.

There are more fancy "multi-tenant" systems such as Salesforce.com where you put a "tenant id" on the database fields so that many tenants can run in a single database, but that's another whole issue in which case some clever thinking would be necessary to add multi-tenant to the open source software.

Oh ok... I kept thinking of the idea of running multiple instances on one machine would be expensive. Thanks for that insight. Since I work with docker stuff a lot I could experiment with this too I guess :) . Still curious what ghost blog's strategy is. Asked them to take a look at this if possible :)

The art here is balancing isolation vs sharing of resources.

If a lot of people share a file system and somebody uploads a lot of big files and fills the disk that can crash the system for everybody.

For a blog service you will have a lot of people who put in 2 or 3 blog posts and never get much traffic, you will have some people who get "lucky" and get 100,000 hits in one day, and a very few who create a lot of content and get lucky a lot.

Shared CPU & memory on a big machine is great when it comes to people getting lucky...

There is no multi-site version of Ghost. Our Pro service runs the same OSS code that we release publicly. We use internal software to setup, manage and administer all the different instances which we run.

There is no difference in features between the open source version and the Pro service.

Whoops with the misunderstanding!!! Thanks so much for replying. Was just wondering though if you'd be up for sharing how that process of managing different instancea is managed. I'm just super interested in learning how others have solved this problem. It might make for an awesome blog post too :)

What you need is minimal 1 machine with a webserver that can host many virtual hosts under 1 or more IP numbers. On that host you run a database server of some kind, most used is MySQL (which can run many concurrent databases). Then each blog will run in a virtual host with a database connection to it's own database. If you use hppts, you need a wildcard ssl certificate that can accommodate all virtual hosts.

This model can be up-scaled to separate machines for the webserver and database server, or even N number of machines running the webserver and N machines running the database server all behind a loadbalancer. All this even under different IP addresses, or usage of CDN services.

In theory you can upscale this the size of Google services if you have deep pockets.

Short answer, it's not the open source package but the infrastructure behind it, that defines the scale (of course efficient source code will help run things smoother).

Reply on news.ycombinator.com