Gipeda -- the Git Performance Dashboard

What is gipeda?

Gipeda is a a tool that presents data from your program’s benchmark suite (or any other source), with nice tables and shiny graphs.

It is only a frontend and does not help with or care about collecting the data. So it is up to you whether you have a polling shell script loop, a post-commit hook or a elaborate jenkins setup. As long as the performance data ends up in the logs/ directory, gipeda is happy.

Gipeda produces static pages. In fact, the (single) html file and the accompagning JavaScript code is completely static. Gipeda just generates a large number of json files. This has the advantage of easy deployment: Just put gipeda in your webspace of copy the files to some static web hosting and you are done. This putts very little load on your server, is cache friendly and has no security problems.

Do you want to see it live? Check out these:

• Demo page, visualizing fairly boring stuff about gipeda itself.
• GHC’s gipeda installation.

Setting it up

• Clone gipeda somewhere, possibly directly into your webspace.

• Install a Haskell compiler, including the cabal tool.

• Install a few packages

   apt-get install git unzip libssl-dev libfile-slurp-perl libipc-run-perl libicu-dev
  

• Install the dependencies:

   cabal install --only-dependencies
  

• Compile it:

   cabal install --bindir=.
  

• Create a gipeda.yaml. You can look at the example file.

• Clone the repository of your project into repository/. A bare clone is sufficient, e.g.

   git clone  --bare git://git.haskell.org/ghc.git repository
  

• Download a bunch of JavaScript libraries by running ./install-jslibs.sh.

Gipeda does not work without at least some logs, so lets add them.

Adding data

Gipeda expect simple CSV files for each revision, of the form

benchmark1;1000
benchmark2;20.123
benchmark3;0

But likely your benchmark suite does not generate them in this format directly. Hence, put whatever format you have (text base logs, JUnit reports, whatever) into the directory logs, named <gitrev>.log, e.g. logs/0279a7d327a3b962ffa93a95d47ea5d9ee31e25c.log.

Then create a script log2csv that expects the filename of such a log on the command line and produces the desired CSV file.

Running gipeda

With everything in place, you can now run

./gipeda

and it will create a bunch of JSON files in site/out/. With ./gipda -j4 you can parallelize it.

You should do this everytime a new log file appears in logs/. You should also make sure your repository is up-to-date, e.g. by running git -C repository pull or, if it is a bare clone, git -C repository fetch origin "+refs/heads/*:refs/heads/*" --prune.

Using gipeda

Finally, you simply point your browser to the site/index.html. The page should be mostly self-explanatory. If you don’t see anything, it might be because of the filter in the top-right corner. Try to enable all buttons, even the =.

To host this on a webserver, just put the site/ directory in your webspace.

Hacking on gipeda

Gipeda doesn't do much; it mostly assembles the data and creates nice reports. The rough pipeline is as follows:

• Directory logs/ contains project-specific data per git commit that has been benchmarked. gipeda will run log2csv on these files to generate the files in site/out/results. logs may be a normal directory, or (for disk space efficiency) a bare git repository. This step is optional.

• Directory site/out/results contains one csv file per git commit. The format is simple, as there are two columns: benchmark name and a numerical value.

• From these files, gipeda generates a number of JSON files, some per commit (report, summaries), some global (settings, latest-summaries).

  A crucial idea here is that these JSON files are all but fragments of a theoretical global JSON document. In other words: You could combine them (using a naive JSON object merge) and there would be no conflicts, and the result could be used by the client as well.

• The client (site/index.html and site/js/gipeda.js) is a fairly standard HTML+JS application using jquery, bootstrap, handlebars.

Bugs, Code, Contact

Please reports bugs and missing features at the GitHub bugtracker. This is also where you can find the source code.

Gipeda was written by Joachim Breitner and is licensed under a permissive MIT license.