Thor is WebSocket benchmarking/load generator. There are a lot of benchmarking tools for HTTP servers. You've got ab, siege, wrk and more. But all these tools only work with plain ol HTTP and have no support for WebSockets - even if they did they wouldn't be suitable, as they would be testing short running HTTP requests instead of long running HTTP requests with a lot of messaging traffic. Thor fixes all of this.
Thor requires Node.js to be installed on your system. If you don't have Node.js installed you can download it from http://nodejs.org or build it from the github source repository: https://github.com/joyent/node.
Once you have Node.js installed, you can use the bundled package manager npm
to
install this module:
npm install -g thor
The -g
command flag tells npm
to install the module globally on your system.
thor [options] <urls>
Thor can hit multiple URL's at once; this is useful if you are testing your
reverse proxies, load balancers or just simply multiple applications. The url
that you supply to thor
should be written in a WebSocket compatible format
using the ws
or wss
protocols:
thor --amount 5000 ws://localhost:8080 wss://localhost:8081
The snippet above will open up 5000
connections against the regular
ws://localhost:8080
and also 5000
connections against the secured
wss://localhost:8081
server, so a total of 10000
connections will be made.
One thing to keep in mind is you probably need to bump the amount of file
descriptors on your local machine if you start testing WebSockets. Set the
ulimit -n
on machine as high as possible. If you do not know how to do this,
Google it.
Usage: thor [options] ws://localhost
Options:
-h, --help output usage information
-A, --amount <connections> the amount of persistent connections to generate
-C, --concurrent <connections> how many concurrent-connections per second
-M, --messages <messages> messages to be send per connection
-P, --protocol <protocol> WebSocket protocol version
-B, --buffer <size> size of the messages that are send
-W, --workers <cpus> workers to be spawned
-G, --generator <file> custom message generators
-M, --masked send the messaged with a mask
-b, --binary send binary messages instead of utf-8
-V, --version output the version number
Some small notes about the options:
--protocol
is the protocol version number. If you want to use the HyBi drafts 07-12 use8
as argument or if you want to use the HyBi drafts 13-17 drafts which are the default version use13
.--buffer
should be size of the message in bytes.--workers
as Node.js is single threaded this sets the amount of sub processes to handle all the heavy lifting.
Some WebSocket servers have their own custom messaging protocol. In order to
work with those servers we introduced a concept called generators
a generator
is a small JavaScript file that can output utf8
and binary
messages. It uses
a really simple generator by default.
Checkout https://github.com/observing/thor/blob/master/generator.js for an example of a generator.
thor --amount 1000 --generator <file.js> ws://localhost:8080
thor --amount 1000 --messages 100 ws://localhost:8080
This will hit the WebSocket server that runs on localhost:8080 with 1000
connections and sends 100 messages over each established connection. Once thor
is done with smashing your connections it will generate a detailed report:
Thor: version: 1.0.0
God of Thunder, son of Odin and smasher of WebSockets!
Thou shall:
- Spawn 4 workers.
- Create all the concurrent/parallel connections.
- Smash 1000 connections with the mighty Mjölnir.
The answers you seek shall be yours, once I claim what is mine.
Connecting to ws://localhost:8080
Opened 100 connections
Opened 200 connections
Opened 300 connections
Opened 400 connections
Opened 500 connections
Opened 600 connections
Opened 700 connections
Opened 800 connections
Opened 900 connections
Opened 1000 connections
Online 15000 milliseconds
Time taken 31775 milliseconds
Connected 1000
Disconnected 0
Failed 0
Total transferred 120.46MB
Total received 120.43MB
Durations (ms):
min mean stddev median max
Handshaking 217 5036 4094 3902 14451
Latency 0 215 104 205 701
Percentile (ms):
50% 66% 75% 80% 90% 95% 98% 98% 100%
Handshaking 3902 6425 8273 9141 11409 12904 13382 13945 14451
Latency 205 246 266 288 371 413 437 443 701
MIT