As some might realize I have been getting into gRPC as of late for internal API development at the company I work at. In the blog aforementioned I talked about how easy it was to get started with that in golang.
Yesterday, we are concluding a big project where we need to migrate much data at work, and we decided to try gRPC on it because we had this feeling that it was going to be faster than REST calls to our existing API. When we retrospected about the project yesterday it was mentioned that we could have finished the project faster if maybe we did the project in a technology that more people on the team knew, basically if we didn’t do gRPC more people could swarm on that project.
Here I am going to do some benchmarks against a classic REST API and the exact same endpoint in gRPC so we can see the performance differences.
Results for the impatient
I wrote some code to show the amazing difference. Here are the results:
Wow 25 Times performance gain!!
Setup for the interested
In order to do an apples to apples comparison, I created a REST HTTPS endpoint that
took in json, performed a validation check on the input, and then responded.
Along the same lines, I also created a ProtoBuffer service description, converted that to golang code and copy and pasted the same validation mechanism.
Here is the REST http.HandlerFunc:
Pretty reasonable handler. Here is the competing GRPC handler:
Wow, that GRPC endpoint is so much cleaner in my opinion. There is no messy input/response serialization baloney as grpc deals with the serialization. Moreover, there is actually a service description that generates all the InfoRequest/InfoReply structures for me. It is pretty clear that a scheme change would cause many many touch points in the REST version of the code, whereas there are basically no refactoring touch points other than the service definition file.
It is fairly clear to me based on the above that for internal only APIs I much prefer to use gRPC over REST.