gRPC or REST

-

 





I have designed and implemented many REST API's over years. If I have to do them now I would have chosen gRPC for more than half of those scenario's. REST (REpresentational state transfer) was developed early 2000's. It took off and replaced SOAP API's quickly. While gRPC is developed pretty recently in 2015.

A quick comparison

gRPC

REST

Uses HTTP/2

Uses HTTP/1.1

Protocol Buffer

JSON or XML

Limited browser support through proxy layer

Universal browser support

Client - response model. 

Bidirectional support.

Request - response model

7x faster



If you do not consume the API's directly in your browser you should go gRPC. If your use case needs your browser to make API calls then REST is the option for you. I do not have the performance numbers of gRPC with a proxy like Envoy. I will update the article soon with that.

All other aspects are comparable and give advantage to gRPC. If you want more details these are my references:


Comments

Post a Comment

Popular posts from this blog

ChatGPT for cavemen

-
Image
I heard a lot about ChatGPT and thought to come out of my cave to find out what the hype is about. Is it going to eat all my lunch or eat me for lunch?  Like any product I gave it all my details and signed up. I asked it about quantum computing an it gave some answer. So far so good. Next day I wanted to check its blogging skills. So I opened chatGPT and I got this . I checked throughout the day and it is still down. So it won’t eat me till it is fed few more servers.  Continuing after it was up.  I asked it to write a blog about itself. Here is what was generated by chatGPT.
Read more

Greedy computing with API

-
Image
 We all know computing resources have a cost. Memory, long term storage, transportation and computing all have their own costs. There are lot of scenarios where your API end point won't be able to respond in time. These are scenarios where you need to use more resources to provide a response. Let us be greedy with our resources and explore different patterns which can solve this. Polling  This is a commonly used pattern where the caller initiates a process. Then keeps on checking if the process is complete. If the process is complete then the caller reads the data needed.  This is a good pattern when both systems are part of the same organization. At least developers of caller and callee should be able to discuss this implementation.  Call back This is an improvement over polling pattern. Here the caller supplies a call back URL to the long running process. Once the process is complete it calls the callback URL and informs the caller that process results are available now. 
Read more

Event driven architecture for micro services

-
This is a work in progress. I am designing an event driven architecture to be used in systems using micro services. Micro service architecture is popular because it solves few problems in development process. While it solves some problems it also brings in new problems to the table. Bad designs impact micro services harder than monoliths. But this is thought is for another time.  How micro services talk to each other. In a monolith you have one gigantic database, or you can easily call another component to process the information. In a microservice you can use a Que (or event bus) to pass the information. This is the problem I want to discuss. Main issue with que is it is not designed for this scenario. There are few products which give a good solution by providing a subscription-based approach. But these are large implementations. Using these is similar to bringing an excavator to plant a sapling in your garden. Yes, many people go with using these large implementations to resolve thi
Read more