# Deployment This document describes the necessary steps to configure patchwork in a production environment. This requires a significantly "harder" deployment than the one used for development. If you are interested in developing patchwork, please refer to [the development guide][doc-development] instead. This document describes a two-node installation of patchwork, consisting of a database sever and an application server. It should be possible to combine these machines for smaller patchwork instances. It should also be possible to configure high availability deployment through use of additional database and application machines, though this is out of the scope of this document. ## Deployment Guides, Provisioning Tools and Platform-as-a-Service Before continuing, it's worth noting that patchwork is a Django application. With the exception of the handling of incoming mail (described below), it can be deployed like any other Django application. This means there are tens, if not hundreds, of existing articles and blogs detailing how to deploy an application like this. As such, if any of the below information is unclear then we'd suggest you go search for "Django deployment guide" or similar, deploy your application, and submit [a patch for this guide][doc-contributing] to clear up that confusion for others. You'll also find that the same search reveals a significant number of existing deployment tools aimed at Django. These tools, be they written in Ansible, Puppet, Chef or something else entirely, can be used to avoid much of the manual configuration described below. If possible, embrace these tools to make your life easier. Finally, many Platform-as-a-Service (PaaS) providers and tools support deployment of Django applications with minimal effort. Should you wish to avoid much of the manual configuration, we suggest you investigate the many options available to find one that best suits your requirements. The only issue here will likely be the handling of incoming mail - something which many of these providers don't support. We address this in the appropriate section below. ## Requirements For the purpose of this guide, we will assume the following machines: | server role | IP address | |-------------|------------| | database | 10.1.1.1 | | application | 10.1.1.2 | We will use the database server to, ostensibly enough, host the database for the patchwork instance. The application server, on the other hand, will host the patchwork instance along with the required reverse proxy and WSGI HTTP servers. We expect a Ubuntu 15.04 installation on each of these hosts: commands, package names and/or package versions will likely change if using a different distro or release. In addition, usage of different package versions to the ones suggested may require slightly different configuration. Before beginning, you should update these systems: $ sudo apt-get update $ sudo apt-get upgrade We also need to configure some environment variables to ease deployment. These should be exported on all systems:
PW_HOST_DB=10.1.1.1
IP of the database host
PW_HOST_APP=10.1.1.2
IP of the application host
PW_DB_NAME=patchwork
Name of the database
PW_DB_USER=www-data
Username that the patchwork app will access the database with
## Database These steps should be run on the database server. **NOTE:** If you already have a database server on site, you can skip much of this section. ### Install Requirements We're going to rely on PostgreSQL. You can adjust the below steps if using a different RDBMS. Install the required packages. $ sudo apt-get install -y postgresql postgresql-contrib ### Configure Database PostgreSQL created a user account called `postgres`; you will need to run commands as this user. Use this account to create the database that patchwork will use, using the credentials we configured earlier. $ sudo -u postgres createdb $PW_DB_NAME $ sudo -u postgres createuser $PW_DB_USER We will also need to apply permissions to the tables in this database but seeing as the tables haven't actually been created yet this will have to be done later. **TODO** `pg_hba.conf` configuration ## Patchwork These steps should be run on the application server. ### Install Packages The first requirement is patchwork itself. It can be downloaded like so: $ wget https://github.com/getpatchwork/patchwork/archive/v1.1.0.tar.gz We will install this under `/opt`, though this is only a suggestion: $ tar -xvzf v1.1.0.tar.gz $ sudo mv v1.1.0 /opt/patchwork **NOTE:** Per the [Django documentation][ref-django-files], source code should not be placed in your web server's document root as this risks the possibility that people may be able to view your code over the Web. This is a security risk. Next we require Python. If not already installed, then you should do so now. patchwork supports both Python 2.7 and Python 3.3+, though we would suggest using the latter to ease future upgrades: $ sudo apt-get install python3 # or 'python' if using Python 2.7 We require a number of Python packages. These can be installed using `pip`: $ sudo pip install -r /opt/patchwork/requirements-prod.txt If you're not using `pip`, you will need to identify and install the corresponding distro package for each of these requirements. For example: $ sudo apt-get install python3-django **NOTE:** The [pkgs.org][ref-pkgs] website provides a great reference for identifying the name of these dependencies. ### Configure patchwork You will also need to configure a [settings][ref-django-settings] file for Django. A sample settings file is provided that defines default settings for patchwork. You'll need to configure settings for your own setup and save this as `production.py`. $ cp patchwork/settings/production.example.py \ patchwork/settings/production.py Alternatively, you can override the `DJANGO_SETTINGS_MODULE` environment variable and provide a completely custom settings file. **NOTE:** You should not include shell variables in settings but rather hardcoded values. These settings files are evaluated in Python - not a shell. ### Databases You can configure the `DATABASES` setting using the variables we set earlier. DATABASES = { 'default': { 'ENGINE': 'django.db.backends.postgresql_psycopg2', 'HOST': '$PW_HOST_DB', # don't use sh variables but actual values 'PORT': '', 'NAME': '$PW_DB_NAME', 'USER': '$PW_DB_USER', 'PASSWORD': '$PW_DB_PASS', 'TEST': { 'CHARSET': 'utf8', }, }, } **NOTE:** `TEST/CHARSET` is used when creating tables for the test suite. Without it, tests checking for the correct handling of non-ASCII characters fail. It is not necessary if you don't plan to run tests, however. #### Static Files While we have not yet configured our proxy server, we do need to configure the location that these files will be stored in. We will install these under `/var/www/patchwork`, though this is only a suggestion and can be changed. $ mkdir /var/www/patchwork You can configure this by setting the `STATIC_ROOT` variable. STATIC_ROOT = '/var/www/patchwork' #### Other Options Finally, the following settings need to be configured. The purpose of these variables is described in the [Django documentation][ref-django-settings]: * `SECRET_KEY` * `ADMINS` * `TIME_ZONE` * `LANGUAGE_CODE` * `DEFAULT_FROM_EMAIL` * `NOTIFICATION_FROM_EMAIL` You can generate the `SECRET_KEY` with the following python code: import string, random chars = string.letters + string.digits + string.punctuation print repr("".join([random.choice(chars) for i in range(0,50)])) If you wish to enable the XML-RPC interface, you should add the following to the file: ENABLE_XMLRPC = True ### Final Steps Once done, we should be able to check that all requirements are met using the the `check` command of the `manage.py` executable: $ /opt/patchwork/manage.py check We should also take this opportunity to both configure the database and static files: $ /opt/patchwork/manage.py migrate $ /opt/patchwork/manage.py loaddata \ /opt/patchwork/patchwork/fixtures/default_tags.xml $ /opt/patchwork/manage.py loaddata \ /opt/patchwork/patchwork/fixtures/default_states.xml $ /opt/patchwork/manage.py collectstatic **NOTE:** The above `default_tags` and `default_states` are just that: defaults. You can modify these to fit your own requirements. Finally, it may be helpful to start the development server quickly to ensure you can see *something*: $ /opt/patchwork/manage.py runserver 0.0.0.0:8080 Browse this instance at `http://[your_server_ip]:8000`. If everything is working, kill the development server using `Ctrl`+`C`. ## Reverse Proxy and WSGI HTTP Servers These steps should be run on the application server. ### Install Packages We will use nginx and uWSGI to deploy patchwork, acting as reverse proxy server and WSGI HTTP server respectively. Other options are available, such as Apache+mod_wsgi or nginx+Gunicorn. While we don't document these, sample configuration files for the former case are provided in `lib/apache2/`. $ sudo apt-get install nginx-full uwsgi uwsgi-plugin-python ### Configure nginx and uWSGI Configuration files for nginx and uWSGI are provided in the `lib` subdirectory of the patchwork source code. These can be modified as necessary, but for now we will simply copy them. First, let's load the provided configuration for nginx: $ sudo cp /opt/patchwork/lib/nginx/patchwork.conf \ /etc/nginx/sites-available/ If you wish to modify this configuration, now is the time to do so. Once done, validate and enable your configuration: $ sudo nginx -t $ sudo ln -s /etc/nginx/sites-available/patchwork.conf \ /etc/nginx/sites-enabled/patchwork.conf Now use the provided configuration for uWSGI: $ sudo mkdir -p /etc/uwsgi/sites $ sudo cp /opt/patchwork/lib/uwsgi/patchwork.ini \ /etc/uwsgi/sites/patchwork.ini **NOTE** We created the `/etc/uwsgi` directory above because we're going to run uWSGI in ["emperor mode][ref-uwsgi-emperor]". This has benefits for multi-app deployments. ### Create systemd Unit File As things stand, uWSGI will need to be started manually every time the system boots, in addition to any time it may fail. We can automate this process using systemd. To this end a [systemd unit file][ref-uwsgi-systemd] should be created to start uWSGI at boot: $ sudo cat << EOF > /etc/systemd/system/uwsgi.service [Unit] Description=uWSGI Emperor service [Service] ExecStartPre=/usr/bin/bash -c 'mkdir -p /run/uwsgi; chown user:nginx /run/uwsgi' ExecStart=/usr/bin/uwsgi --emperor /etc/uwsgi/sites Restart=always KillSignal=SIGQUIT Type=notify NotifyAccess=all [Install] WantedBy=multi-user.target EOF **NOTE:** On older version of Ubuntu you may need to tweak these steps to use [upstart][ref-uwsgi-upstart] instead. ### Final Steps Start the uWSGI service we created above: $ sudo systemctl uwsgi start $ sudo systemctl uwsgi status Next up, restart the nginx service: $ sudo systemctl nginx restart $ sudo systemctl nginx status patchwork uses a cron script to clean up expired registrations and send notifications of patch changes (for projects with this enabled). Something like this in your crontab should work. # m h dom mon dow command */10 * * * * cd patchwork; ./manage.py cron **NOTE**: The frequency should be the same as the `NOTIFICATION_DELAY_MINUTES` setting, which defaults to 10 minutes. Finally, browse to the instance using your browser of choice. You may wish to take this opportunity to setup your projects and configure your website address (in the Sites section of the admin console, found at `/admin`). ## Incoming Email patchwork is designed to parse incoming mails which means you need an address to receive email at. This is a problem that has been solved for many webapps, thus there are many ways to go about this. Some of these ways are discussed below. ### Postfix The most flexible option is to configure our own mail transfer agent (MTA) and Postfix is as good a choice as any. While we don't cover setting up Postfix here (it's complicated and there are many guides already available), patchwork does include a script to take received mails and create the relevant entries in patchwork for you. To use this, you should configure your system to forward all emails to a given localpart (the bit before the `@`) to this script. Using the `patchwork` localpart (e.g. `patchwork@example.com`) you can do this like so: $ sudo cat << EOF > /etc/aliases patchwork: "|/opt/patchwork/patchwork/bin/parsemail.sh" EOF You should ensure the appropriate user is created in PostgreSQL and that it has (minimal) access to the database. Patchwork provides scripts for the latter and they can be loaded as seen below: $ sudo -u postgres createuser nobody $ sudo -u postgre psql -f \ /opt/patchwork/lib/sql/grant-all.postgres.sql patchwork **NOTE:** This assumes your Postfix process is running as the `nobody` user. If this is not correct (use of `postfix` user is also common), you should change both the username in the `createuser` command above and substitute the username in the the `grant-all-postgres.sql` script with the appropriate alternative. ### IMAP/POP3 One could also use an email account provided by a run-of-the-mill email provider and retrieve mail using IMAP or POP3. This may be suitable for low-volume mailing lists but be warned: this will introduce a significant lag between when a patch is submitted to a mailing list and when it appears in patchwork. ### Use a Email-as-a-Service Provider Setting up an email server can be a difficult task and, in the case of deployment on PaaS provider, may not even be an option. In this case, there are a variety of web services available that offer "Email-as-as-Service". These services typically convert received emails into HTTP POST requests to your endpoint of choice, allowing you to sidestep configuration issues. We don't cover this here, but a simple wrapper script coupled with one of these services can be more than to get email into patchwork. You can also create such as service yourself using a PaaS provider that supports incoming mail and writing a little web app. ## (Optional) Configure your VCS to Automatically Update Patches The `tools` directory of the patchwork distribution contains a file named `post-receive.hook` which is a sample git hook that can be used to automatically update patches to the `Accepted` state when corresponding commits are pushed via git. To install this hook, simply copy it to the `.git/hooks` directory on your server, name it `post-receive`, and make it executable. This sample hook has support to update patches to different states depending on which branch is being pushed to. See the `STATE_MAP` setting in that file. If you are using a system other than git, you can likely write a similar hook using `pwclient` to update patch state. If you do write one, please contribute it. [doc-contributing]: contributing.md [doc-development]: development.md [ref-django-files]: https://docs.djangoproject.com/en/dev/intro/tutorial01/#creating-a-project [ref-django-settings]: https://docs.djangoproject.com/en/1.8/ref/settings/ [ref-pkgs]: http://pkgs.org/ [ref-uwsgi-emperor]: https://uwsgi-docs.readthedocs.org/en/latest/Emperor.html [ref-uwsgi-systemd]: https://uwsgi-docs.readthedocs.org/en/latest/Systemd.html [ref-uwsgi-upstart]: https://uwsgi-docs.readthedocs.org/en/latest/Upstart.html