DEPLOYMENT.md

How to deploy BrowserID

This describes how to take the code here, put it on a server, and build a service like browserid.org.

So what are we deploying, anyway?

• the browserid server - a node.js server which implements a web services api, stores a record of users, the email addresses they've verified, a bcrypted password, outstanding verification tokens, etc.
• the verifier - a stateless node.js server which does cryptographic verification of assertions. This thing is hosted on browserid.org as a convenience, but people using browserid can choose to relocated it if they want to their own servers.
• the browserid.org website - the templates, css, and javascript that make up the visible part of browserid.org
• the javascript/HTML dialog & include library - this is include.js and the code that it includes, the bit that someone using browserid will include.

Overview

Software in use

This document assumes we're deploying on an Ubuntu 10.04.1 LTS box, and using the following software:

• nginx - frontend web server that handles static content and serves as a reverse proxy for node.js servers running on localhost

  config: /etc/nginx/conf/nginx.conf

• node.js - all non-static servers run with node. modules are installed using npm in /home/http/node_modules

• monit - provides monitoring and automatic restarting of node.js servers when they go down (by accident or upon code publishing).

  config files are: /etc/monitrc, and /etc/monit.d/*

  helper script that starts node servers: /etc/monit.d/start_node_server

• gitolite - installed under the git user to provide multi-user ssh based git access. post-update hook handles updating code and restarting servers.

  hook: /home/git/.gitolite/hooks/common/post-update

Permissions conventions

• nginx runs as user 'www-data'
• node.js servers run as user 'www-data'
• when git pushing, all publishing and restarting runs as user 'git'

Setup

1. Install gitolite

This step is optional. gitlite turns a normal unix machine into a "git server". All that gitolite does is provide some utilities and the infrastructure required to make it possible for multiple users to authenticate to a particular user on the box using ssh keys for the purposes of updating code. While requiring a bit of setup, in practice this is a fabulously lightweight way to make the releases process sing.

Let's get started:

1. create a "git" user: sudo adduser git
2. install git if required: sudo apt-get install git-core
3. become user git: sudo su -s /bin/bash git
4. hop into your home directory: cd
5. install gitolite: This.
6. add a browserid repo: This.

At this point you've morphed your servers into git servers. Go ahead and add a remote to your local copy of the browserid repo and push to it: git remote add shortaliasforthenewvm git@myserver:browserid.git && git push --all shortaliasforthenewvm

Now you have a clone of your browserid repository that's trivial to update. You can use ssh keys with passphrases and ssh-agent if security is a concern.

2. Install node.js

At present we're running node.js 0.4.10. Lastest along the 4 line should work:

1. install dev tools, required to build node: apt-get install g++ libssl-dev
2. build and install node: ./configure && make && sudo make install
3. install npm, required to install uglify-js: git clone https://github.com/isaacs/npm.git && cd npm && sudo make install
4. intstall uglify-js, required to create production resources: npm install -g uglify-js

3. Install software prerequisites

Subsequent steps use different software which you might need to install.

• curl - used to initiate http requests from the command line (to kick the browserid server)
• java - used to minify css
• mysql 5.1+ - the preferred persistence backend

4. Set up mysql

0. ensure you can connect via TCP - localhost:3306 (like, make sure skip-networking is off in my.cnf)
1. connect to the database as user root - mysql -u root
2. CREATE USER 'browserid'@'localhost' IDENTIFIED BY 'browserid';
3. CREATE DATABASE browserid;
4. GRANT CREATE, DELETE, INDEX, INSERT, LOCK TABLES, SELECT, UPDATE ON browserid.* TO 'browserid'@'localhost';

5. Set up git post-update hook

This step is optional - if you want to manually update code, then you probably skipped step #1; you can skip this one as well. All you need to do is check out the code from github and run node.

Given that we've now got a simple way to push updates to the server, and we've got npm and node running, let's get the software running! The task here is to create a git post-update hook to have the server update its code and restart the server when code is pushed.

To get this done, we'll configure a hook which will live on your server in the git user's git directory:

First, do this to add a blank executable post-update hook.

Now, here's a full sample script that you can start with for that post-update hook, annotated to help you follow along:

#!/bin/bash

# only run these commands if it's the browserid repo bein' pushed
if [ "x$GL_REPO" == 'xbrowserid' ] ; then
    # create a temporary directory where we'll stage new code
    NEWCODE=`mktemp -d`
    echo "staging code to $NEWCODE"
    mkdir -p $NEWCODE
    git archive --format=tar dev | tar -x -C $NEWCODE

    echo "generating production resources"
    cd $NEWCODE/browserid && ./compress.sh && cd -

    # stop the servers
    curl -o --url http://localhost:62700/code_update > /dev/null 2>&1
    curl -o --url http://localhost:62800/code_update > /dev/null 2>&1

    # now move code into place, and keep a backup of the last code
    # that was in production in .old
    echo "moving updated code into place"
    rm -rf /home/browserid/code.old
    mv /home/browserid/code{,.old}
    mv $NEWCODE /home/browserid/code

    echo "fixing permissions"
    find /home/browserid/code -exec chgrp www-data {} \; > /dev/null 2>&1
    find /home/browserid/code -type d -exec chmod 0775 {} \; > /dev/null 2>&1
    find /home/browserid/code -type f -exec chmod ga+r {} \; > /dev/null 2>&1
    find /home/browserid/code -type f -perm /u+x -exec chmod g+x {} \; > /dev/null 2>&1

    echo "updating dependencies"
    ln -s /home/browserid/node_modules /home/browserid/code/node_modules
    cd /home/browserid/code && npm install && cd -
fi

6. Get node servers running

At this point, pushing code to gitolite will cause /home/browserid/code to be updated. Now we need to get the servers running! First, verify that the servers will run manually. For the browserid server:

cd /home/browserid/code/browserid && sudo -u www-data ./run.js

And for the verifier:

cd /home/browserid/code/verifier && sudo -u www-data ./run.js

Now let's set up [monit] to restart the node.js servers:

1. install monit: sudo apt-get install monit
2. enable monit by editing /etc/default/monit
3. configure monit. make /etc/monit/monitrc look like this:

set daemon 10
set logfile /var/log/monit.log
include /etc/monit.d/*

4. Add a little utility script (chmod +x) to run the node servers at /etc/monit/start_node_server:

#!/bin/bash
/usr/local/bin/node $1 > $(dirname $1)/error.log 2>&1 &

5. create a file to run the verifier at /etc/monit.d/verifier:

check host verifier with address 127.0.0.1
    start program = "/etc/monit/start_node_server /home/browserid/code/verifier/run.js"
        as uid "www-data" and gid "www-data"
    stop program  = "/usr/bin/pkill -f '/usr/local/bin/node /home/browserid/code/verifier/run.js'"
    if failed port 62800 protocol HTTP
        request /__heartbeat__
        with timeout 10 seconds
        then restart

5. create a file to run the browserid server at /etc/monit.d/browserid:

check host browserid.org with address 127.0.0.1
    start program = "/etc/monit/start_node_server /home/browserid/code/browserid/run.js"
        as uid "www-data" and gid "www-data"
    stop program  = "/usr/bin/pkill -f '/usr/local/bin/node /home/browserid/code/browserid/run.js'"
    if failed port 62700 protocol HTTP
        request /__heartbeat__
        with timeout 10 seconds
        then restart

6. verify servers are running! check /var/log/monit.log, curl ports 62700 and 62800, and verify servers are restarted after 10 seconds if you kill em!

7. Install nginx

At this point we've got automatic server restart, simple git based code publishing, and all of the software prerequisites installed on the box. The final bit of work is to set up nginx in such a way that it will properly proxy requests to the external interface to the proper node server:

1. remove any other webservers that come with your vm (like apache)
2. install nginx: sudo apt-get install nginx
3. configure nginx, make /etc/nginx/nginx.conf look like this:

user www-data;
worker_processes  1;

error_log  /var/log/nginx/error.log;
pid        /var/run/nginx.pid;

events {
    worker_connections  1024;
    # multi_accept on;
}

http {
    include       /etc/nginx/mime.types;
    default_type  application/octet-stream;
    access_log    /var/log/nginx/access.log;
    sendfile        on;

    keepalive_timeout  65;
    tcp_nodelay        on;

    gzip  on;
    gzip_disable "MSIE [1-6]\.(?!.*SV1)";
    gzip_proxied any;
    gzip_types text/html application/json application/javascript text/css application/x-font-ttf application/atom+xml;

    include /etc/nginx/conf.d/*.conf;
    include /etc/nginx/sites-enabled/*;
} 

4. and how about configuring the webserver:

server {
    listen       80 default;
    server_name  browserid.org;

    # disallow external server restart.
    location = /code_update {
        internal;
    }

    # pass /verify invocations to the verifier
    location /verify {
        proxy_pass        http://127.0.0.1:62800;
        proxy_set_header  X-Real-IP  $remote_addr;
    }

    # pass everything else the browserid server
    location / {
        proxy_pass        http://127.0.0.1:62700;
        proxy_set_header  X-Real-IP  $remote_addr;
    }
}

5. restart your webserver: `sudo /etc/init.d/nginx restart

8. Set up log rotation

create a file as root at /etc/logrotate.d/browserid:

/home/browserid/var/log/browserid.log /home/browserid/var/log/browserid-metrics.json /home/browserid/var/log/verifier.log /home/browserid/var/log/verifier-metrics.json {
    daily
    rotate 30
    copytruncate
    dateext
    compress
    dateformat -%Y-%m-%d
}

Now your logfiles will be automatically rotated.

Easy, right?