npmbench
Install & Usage
npm install npmbench
Example: benchmarking the redis package, and running the file multi_bench.js against each release:
npmbench redis multi_bench.js Ops/sec***** v0.0.1******* v0.0.2******** v0.0.3**** v0.0.4********** v0.0.5A command line tool to examine your code's performance over various releases.
- Downloads all releases
- runs the specified js file against all versions
- shows you a graph of your module's performance (Note: this may or may not work depending on how your javascript file prints its results to stdout).
Gotchas
- specialized to read ops/sec from stdout
- releases with incompatible APIs will likely throw errors when the js file is run, and will be omitted aka set to 0. If the error is more serious, things may just crash and I'm sorry.
Your bench should output something like this (even just one line that says 888 ops/sec will work just fine):
[... whatever test output you like ...] 999.9 ops/sec
[... whatever test output you like ...] 999.9 ops/sec
[... whatever test output you like ...] 999.9 ops/sec
Summary: there must be at least one line, and the 2nd to last whitespace separated word must be a number representing the ops/sec.
Niceties of npmbench's approach
- Output from each run of your js file will be put into a file named
npmbench-moduleName@x.y.z.txt. It skips running a given bench if there is already a .txt file there for it. - These are then parsed into json files which are put at
npmbench-moduleName@x.y.z.json - makes it easy to run your own analysis of the numbers after all the benches have been run, or just grep through it :)
Why is this nice? Makes it easy to write your own parser for the output.
TODO:
- graph the graph with something fancier than asterisks on the command line.
- split out the parsing stage so people can supply their own output parser (too much work?)
- actually write the thing that aggregates all the json blobs into one graphable json blob