pdf.js-bot

This bot is used by pdf.js reviewers to run regression tests. The bot lives in server.js and requires node.js. See instructions below on how to set it up.

Getting started

To issue bot requests use the commands below as comments in open pull requests (PR). Here's an example:

• mozilla/pdf.js#603

Note that the bot only listens for commands from repo collaborators (auto-whitelist). Currently our tests are taking approximately 25-30 mins (including our very long pdf.pdf test file), and we've implemented a 60-min timeout for tests that for some reason hang.

• @pdfjsbot test : This will run the full suite of tests on the PR, including image comparison tests against the reference repo. In case there are any image differences, a URL/link will be provided in a comment to the pull request to allow reviewers and requesters to inspect the images and their differences.

• @pdfjsbot makeref : This will generate reference images using the pull request source code and push the resulting images to the reference repo. Note that THIS WILL OVERWRITE any existing images, so only do this if you definitely approve all the visual changes introduced in the PR.

The nitty-gritty:

Here are a few things you might want to know, might have to know in the near future when certain WARNING flags come up, or might never need to know simply because you're so lucky and things just work for you!

• Commands are run in a queue : The bot currently doesn't support concurrent tests. Your order in the queue is reported by the bot upon command recognition.

• By default, test files come from upstream : To ensure we are always testing new code against the latest and most comprehensive regression tests, by default the bot checks out test/ from upstream into the PR clone. The exception is when the pull request itself has new/modified files in test/, in which case the bot uses test/ files from the PR and issues a WARNING explaining it's not using upstream for tests. (This fallback is in place to allow PRs to introduce new tests).

  If there's any suspicion that the PR tests will miss important tests from upstream, the reviewer should ask the requester to merge upstream into the PR branch, and run the test again.

• Reference images are not versioned : Because of their size we don't version our images repo. This means that a pull request might get compared against images with new features that are not present in the requester's branch. In this case the bot will issue a WARNING explaining the situation.

  If the tests pass, it shouldn't be a problem. If they don't, the reviewer should ask the requester to merge upstream into their PR and try again, as the regression might be due to a missing commit.

Troubleshooting

Logging into EC2

To log into the server:

$ ssh -i <ssh_key_file.pem> ubuntu@<ec2_machine_address>

The two unknowns above should be known by the bot collaborators.

Log file

This is the first place to take a peek at:

/tmp/bot.log

(TODO: place file in /var/log and configure logrotate)

Updating and restarting bot

There's a script for that:

$ cd pdf.js-bot/
$ ./update

Restarting bot

$ cd pdf.js-bot/
$ forever stop 0
$ forever start -o /tmp/bot.log server.js

And cross your fingers :)

Setting up bot server

Distro and basics

The instructions below assume we'll be deploying on Ubuntu 11.04. See http://alestic.com/ for a list of EC2 images.

1. $ sudo apt-get update
2. Install via apt-get: make, g++, zip

Browser, Xvfb

1. Install via apt-get: firefox, xvfb (necessary as we will run browsers without a display); and fonts xfonts-100dpi, xfonts-75dpi, xfonts-scalable, and xfonts-cyrillic.

2. Test xvfb via $ xvfb-run firefox. Firefox shouldn't bail out with a no-display message.

3. Configure xvfb by appending to /etc/profile:

    /usr/bin/Xvfb :1 1>/dev/null 2>/dev/null &
    export DISPLAY=:1
   

Git, Github

1. Install via apt-get: git

2. Create Github ssh key in ~/.ssh: $ ssh-keygen -t rsa -C "[email protected]"

3. DO NOT ADD AN SSH PASSWORD TO THE KEY. This requires setting up ssh-agent and a password prompt upon every boot.

4. Log into Github as @pdfjsbot, add public ssh key from file ~/.ssh/id_rsa.pub

5. Test ssh key: $ ssh -T [email protected]

6. Set up ssh agent to avoid repeated password entry: $ exec ssh-agent bash, $ ssh-add ~/.ssh/id_rsa

7. Configure git signatures: $ git config --global user.name "pdfjs bot", git config --global user.email "[email protected]"

8. Set Github API credentials by appending to /etc/profile:

    export GITHUB_CREDENTIALS=pdfjsbot:<password_here>
   

9. Run $ /etc/profile, or $ sudo reboot to effect profile changes

Node.js

1. Install nodejs and nodejs-dev from binaries: https://github.com/joyent/node/wiki/Installing-Node.js-via-package-manager
2. Make sure nodejs-dev and g++ are installed before proceeding!
3. Install node.js package manager npm: $ curl http://npmjs.org/install.sh | sudo sh
4. Install global node utility forever: $ sudo npm install -g forever

Bot

1. Clone repo into home dir: $ cd ~; git clone <pdf.js-bot-URL>

2. Install required node packages: $ cd pdf.js-bot; npm install

3. Configure browser manifest file: test-files/browser_manifest.json

4. (If using a local reference repo, i.e. not on Github) Initialize bare reference repo:

    $ cd ~; git init --bare ./pdf.js-ref
    $ cd ~/pdf.js-ref
    $ echo nothing yet > README
    $ git add .; git commit -am 'First'
   

Launch: Production

1. Check if default parameters are OK: config.json
2. Check in particular the repo locations in config.json
3. Start bot with $ forever start -o /tmp/bot.log server.js
4. (You can stop server with $ forever stop 0)

Launch: Staging

(Use this for local tests)

1. Configure parameters for your own tests in: config_staging.json
2. Set environment variable $ export PDFJSBOT_STAGING=yes
3. Start bot with $ node server.js