Original post

Hello folks,

It has been a long time since we talked about performance in Go.

There is no secret that Go has support for a large amount of template syntax nowadays. However, we’ve seen zero articles on how each of them performs. Here we are.

Intro

In this article we will run simple, easy to understand, mini benchmarks to find out which template parser renders faster. Keep in mind that speed is not always the case, you have to choose wisely the template parser to use because their features varies and depending your application’s needs you may be forced to use an under-performant parser instead.

We will compare the following 8 template engines/parsers:

Each benchmark test consists of a template + layout + partial + template data (map). They all have the same response amount of total bytes. Amber, Ace and Pug parsers minifies the template before render, therefore, all other template file’s contents are minified too (no new lines and spaces between html blocks).

As always, my benchmark code is available for everyone but I will not tire you anymore by showing the code for each of them in the article, instead you can navigate here.

Benchmarks

Processor Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz
RAM 15.85 GB
OS Microsoft Windows 10 Pro
Bombardier v1.2.4
Go go1.15.2

Terminology

Name is the name of the template engine.

Reqs/sec is the avg number of total requests could be processed per second (the higher the better).

Latency is the amount of time it takes from when a request is made by the client to the time it takes for the response to get back to that client (the smaller the better).

Throughput is the rate of production or the rate at which data are transferred (the higher the better, it depends from response length (body + headers).

Time To Complete is the total time (in seconds) the test completed (the smaller the better).

Results

📖 Fires 1000000 requests with 125 concurrent clients. It receives HTML response. The server handler sets some template data and renders a template file which consists of a layout and a partial footer.

Name Language Reqs/sec Latency Throughput Time To Complete
Amber Go 125698 0.99ms 44.67MB 7.96s
Blocks Go 123974 1.01ms 43.99MB 8.07s
Django Go 118831 1.05ms 42.17MB 8.41s
Handlebars Go 101214 1.23ms 35.91MB 9.88s
Pug Go 89002 1.40ms 31.81MB 11.24s
Ace Go 64782 1.93ms 22.98MB 15.44s
HTML Go 53918 2.32ms 19.13MB 18.55s
Jet Go 4829 25.88ms 1.71MB 207.07s

And for those who like graphs, here you are:

As we’ve seen above, Amber is the fastest and Jet is the slowest one. Keep in mind that in Jet parser, we cache the templates by ourselves before server ran – its default behavior is to cache them at runtime instead, so it could be even slower natively.

If you want to run the benchmarks by yourself, please read the following guide.

Have fun 🥳 and keep yourself strong in these difficult and anxious times we live as a society💪

Yours,
Gerasimos Maropoulos. Author of the Iris Web Framework