We are strong advocates of the Serverless framework at GorillaStack. We’ve recently talked about our recent open source Serverless HipChat Connect example that utilizes Serverless and AWS Lambda. During the development of the plugin, one thing we liked was that Serverless doesn’t restrict you to one architecture – there are at least 4 serverless architectures to choose from.
Whenever I hear “monolith”, I think of a massive LAMP project with a single, burning hot MySQL database.
(not always the case). The monolith architecture looks something like this in Serverless:
I.e. all requests to go to a single Lambda function,
app.js. Users and games have nothing to do with one another but the application logic for users and games are in the same Lambda function.
We found that the greatest advantage that the monolith had over nanoservices and microservices was speed of deployment. With nanoservices and microservices, you have to deploy multiple copies of dependent
node_modules (with Node.js) and any library code that your functions share which can be slow. With the monolith, it’s a single function deployment to all API endpoints so deployment is faster. On the other hand, how common is it to want to deploy all endpoints…
This architecture in Serverless has similar drawbacks to the monolithic architecture in general:
app.jsis responsible for both users and games. If a bug gets introduced into the users part of the function, it’s more likely that the games part might break too
Microservices in serverless architecture looks something like this:
The advantages of the microservices architecture in Serverless inherits advantages of microservices in general. A couple:
games.js, calls to
users.jsshould carry on working. Of course, maybe GET /users/:id might contact the games microservice to get games that the user plays but if
games.jsare proper microservices then
games.jsfailing and vice versa
games.jsdoesn’t make the
users.jscodebase any more complicated
As mentioned above with the monolith, microservices in Serverless can result in slower deployments.
Nanoservices take microservices to the extreme – one function per endpoint instead of one per resource:
Nanoservices take the advantages of microservices and amplifies them. Separation of concerns is greater –
get_users.js is probably going to be simpler than
users.js that handles everything to do with a user.
Again, similar to microservices but even more so – the number of functions to deploy can get huge so deployment time can increase.
There is nothing to stop Developers taking a hybrid approach to their serverless architectures e.g. a mixture of nanoservices and microservices. In our example, if there was a lot of game functionality, it might make sense to split the functionality into nanoservices but if is less functionality related to users, microservices could be more appropriate:
If you have any examples of how you’re architecting your Serverless projects, tell us about it!