‘API DAYS : A comparison of REST APIs
GopherConAE 2016 talk
Preet Singh
This repository contains code and slides from my GopherConAE 2016 talk. I presented a performance comparison between REST APIs in 3 tech stacks: Python + Tornado, Golang +net/http, Java+SpringBoot. When deciding between different tech stacks in your web app architecture, this comparison and code can serve as a start point to compare the performance of different tech stacks.
Note : This uses Golang 1.4.2 . Ver 1.6+ has major changes.
Golang was by far the fastest on simple requests, running twice as many requests as Java, and handled nearly fifteen times the requests handled by Python.
Simple request metrics
Performance details with database requests was less clear.
Database request metrics:
Java performed best, but it was clear there were multiple factors in play: SpringBoot seemed to be doing a version of connection caching, creating a much larger footprint. Golang’s connection pooling philosophy by contrast, minimizes the resource footprint, placing a limitation on the maximum number of connections in the pool, rather than allow a minimum:
Connection parameters
-
db.setMaxIdleConns : “Please don’t keep more than n connections when they’re not longer in use.”
-
db.setMaxOpenConns : “Allow only max (n) connections to the db”
Other factors included the efficacy of the device drivers - I used MySQL as the database. Performance metrics with say, MongoDB might present an interesting comparison, since Golang’s mgo driver is allegedly far superior.
Tech stack
Go ver. 1.4.2, only net/http • Python 2.7.11, Tornado ver. 4.3 • Java 1.8, Spring Boot ver. 1.3.2 • Ran on my Macbook Pro 2.4 GHz i5, 8 GB RAM • Used wrk for benchmarking
Files + folders included
The keynote and pdf versions of my presentation, and folders for code in the 3 tech stacks.
Go REST APIs
REST APIs in Golang I've used for benchmarking:
- V0 generates a random number, returns it as the JSON'ed response.
- V1 retrieves ~4000 rows from a MySQL DB, sorts them, returns the JSON’ed result.
Python REST APIs
REST APIs in Python+Tornado that I used for benchmarking in my GopherCon DXB '16 talk.
- Ver. 0 generates a random number and returns the JSON'ed response.
- Ver. 1 retrieves ~4000 rows from a MySQL db, sorts them, and returns the JSON'ed response.
Java REST APIs
- Ver. 1 : REST API in Java+SpringBoot that serves a trivial request I used for benchmarking. The request generates a random number, and returns it in the response as a JSON object.
- Ver 2 : REST API in Java+SpringBoot used for benchmarking. This retrieves ~4000 Names from the City table in the world database. It then sorts out the names, and returns them as a JSON'ed response.