7995: Add "dry run" note.

[arvados.git] / doc / install / install-api-server.html.textile.liquid

---
layout: default
navsection: installguide
title: Install the API server
...

h2. Install prerequisites

The Arvados package repository includes an API server package that can help automate much of the deployment.

h3(#install_ruby_and_bundler). Install Ruby and Bundler

{% include 'install_ruby_and_bundler' %}

h2(#install_apiserver). Install API server and dependencies

On a Debian-based system, install the following packages:

<notextile>
<pre><code>~$ <span class="userinput">sudo apt-get install bison build-essential libcurl4-openssl-dev git arvados-api-server</span>
</code></pre>
</notextile>

On a Red Hat-based system, install the following packages:

<notextile>
<pre><code>~$ <span class="userinput">sudo yum install bison make automake gcc gcc-c++ libcurl-devel git arvados-api-server</span>
</code></pre>
</notextile>

{% include 'install_git' %}

h2(#configure). Set up the database

Configure the API server to connect to your database by updating @/etc/arvados/api/database.yml@. Replace the @xxxxxxxx@ database password placeholder with the "password you generated during database setup":install-postgresql.html#api. Be sure to update the @production@ section.

<notextile>
<pre><code>~$ <span class="userinput">editor /etc/arvados/api/database.yml</span>
</code></pre></notextile>

h2(#configure_application). Configure the API server

Edit @/etc/arvados/api/application.yml@ to configure the settings described in the following sections.  The API server reads both @application.yml@ and its own @config/application.default.yml@ file.  The settings in @application.yml@ take precedence over the defaults that are defined in @config/application.default.yml@.  The @config/application.yml.example@ file is not read by the API server and is provided as a starting template only.

@config/application.default.yml@ documents additional configuration settings not listed here.  You can "view the current source version":https://dev.arvados.org/projects/arvados/repository/revisions/master/entry/services/api/config/application.default.yml for reference.

Only put local configuration in @application.yml@.  Do not edit @application.default.yml@.

h3(#uuid_prefix). uuid_prefix

Define your @uuid_prefix@ in @application.yml@ by setting the @uuid_prefix@ field in the section for your environment.  This prefix is used for all database identifiers to identify the record as originating from this site.  It must be exactly 5 lowercase ASCII letters and digits.

Example @application.yml@:

<notextile>
<pre><code>  uuid_prefix: <span class="userinput">zzzzz</span></code></pre>
</notextile>

h3. secret_token

The @secret_token@ is used for for signing cookies.  IMPORTANT: This is a site secret. It should be at least 50 characters.  Generate a random value and set it in @application.yml@:

<notextile>
<pre><code>~$ <span class="userinput">ruby -e 'puts rand(2**400).to_s(36)'</span>
yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy
</code></pre></notextile>

Example @application.yml@:

<notextile>
<pre><code>  secret_token: <span class="userinput">yyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy</span></code></pre>
</notextile>

h3(#blob_signing_key). blob_signing_key

The @blob_signing_key@ is used to enforce access control to Keep blocks.  This same key must be provided to the Keepstore daemons when "installing Keepstore servers.":install-keepstore.html  IMPORTANT: This is a site secret. It should be at least 50 characters.  Generate a random value and set it in @application.yml@:

<notextile>
<pre><code>~$ <span class="userinput">ruby -e 'puts rand(2**400).to_s(36)'</span>
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
</code></pre></notextile>

Example @application.yml@:

<notextile>
<pre><code>  blob_signing_key: <span class="userinput">xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx</span></code></pre>
</notextile>

h3(#omniauth). sso_app_secret, sso_app_id, sso_provider_url

The following settings enable the API server to communicate with the "Single Sign On (SSO) server":install-sso.html to authenticate user log in.

Set @sso_provider_url@ to the base URL where your SSO server is installed.  This should be a URL consisting of the scheme and host (and optionally, port), without a trailing slash.

Set @sso_app_secret@ and @sso_app_id@ to the corresponding values for @app_secret@ and @app_id@ used in the "Create arvados-server client for Single Sign On (SSO)":install-sso.html#client step.

Example @application.yml@:

<notextile>
<pre><code>  sso_app_id: <span class="userinput">arvados-server</span>
  sso_app_secret: <span class="userinput">wwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwww</span>
  sso_provider_url: <span class="userinput">https://sso.example.com</span>
</code></pre>
</notextile>

h3. workbench_address

Set @workbench_address@ to the URL of your workbench application after following "Install Workbench.":install-workbench-app.html

Example @application.yml@:

<notextile>
<pre><code>  workbench_address: <span class="userinput">https://workbench.zzzzz.example.com</span></code></pre>
</notextile>

h3. websocket_address

Set @websocket_address@ to the @wss://@ URL of the API server websocket endpoint after following "Set up Web servers":#set_up.  The path of the default endpoint is @/websocket@.

Example @application.yml@:

<notextile>
<pre><code>  websocket_address: <span class="userinput">wss://ws.zzzzz.example.com</span>/websocket</code></pre>
</notextile>

h3(#git_repositories_dir). git_repositories_dir

The @git_repositories_dir@ setting specifies the directory where user git repositories will be stored.

The git server setup process is covered on "its own page":install-arv-git-httpd.html. For now, create an empty directory in the default location:

<notextile>
<pre><code>~$ <span class="userinput">sudo mkdir -p /var/lib/arvados/git/repositories</span>
</code></pre></notextile>

If you intend to store your git repositories in a different location, specify that location in @application.yml@.

Default setting in @application.default.yml@:

<notextile>
<pre><code>  git_repositories_dir: <span class="userinput">/var/lib/arvados/git/repositories</span>
</code></pre>
</notextile>

h3(#git_internal_dir). git_internal_dir

The @git_internal_dir@ setting specifies the location of Arvados' internal git repository.  By default this is @/var/lib/arvados/internal.git@.  This repository stores git commits that have been used to run Crunch jobs.  It should _not_ be a subdirectory of @git_repositories_dir@.

Example @application.yml@:

<notextile>
<pre><code>  git_internal_dir: <span class="userinput">/var/lib/arvados/internal.git</span>
</code></pre>
</notextile>

h2(#set_up). Set up Web servers

For best performance, we recommend you use Nginx as your Web server front-end, with a Passenger backend for the main API server and a Puma backend for API server Websockets.  To do that:

<notextile>
<ol>
<li><a href="https://www.phusionpassenger.com/library/walkthroughs/deploy/ruby/ownserver/nginx/oss/install_passenger_main.html">Install Nginx and Phusion Passenger</a>.</li>

<li><p>Install runit to supervise the Puma daemon.  {% include 'install_runit' %}<notextile></p></li>

<li><p>Install the script below as the run script for the Puma service, modifying it as directed by the comments.</p>

<pre><code>#!/bin/bash

set -e
exec 2>&1

# Uncomment the line below if you're using RVM.
#source /etc/profile.d/rvm.sh

envdir="`pwd`/env"
mkdir -p "$envdir"
echo ws-only > "$envdir/ARVADOS_WEBSOCKETS"

cd /var/www/arvados-api/current
echo "Starting puma in `pwd`"

# Change arguments below to match your deployment, "webserver-user" and
# "webserver-group" should be changed to the user and group of the web server
# process.  This is typically "www-data:www-data" on Debian systems by default,
# other systems may use different defaults such the name of the web server
# software (for example, "nginx:nginx").
exec chpst -m 1073741824 -u webserver-user:webserver-group -e "$envdir" \
  bundle exec puma -t 0:512 -e production -b tcp://127.0.0.1:8100
</code></pre>
</li>

<li><p>Edit the http section of your Nginx configuration to run the Passenger server, and act as a front-end for both it and Puma.  You might add a block like the following, adding SSL and logging parameters to taste:</p>

<pre><code>server {
  listen 127.0.0.1:8000;
  server_name localhost-api;

  root /var/www/arvados-api/current/public;
  index  index.html index.htm index.php;

  passenger_enabled on;
  # If you're using RVM, uncomment the line below.
  #passenger_ruby /usr/local/rvm/wrappers/default/ruby;

  # This value effectively limits the size of API objects users can
  # create, especially collections.  If you change this, you should
  # also ensure the following settings match it:
  # * `client_max_body_size` in the server section below
  # * `client_max_body_size` in the Workbench Nginx configuration (twice)
  # * `max_request_size` in the API server's application.yml file
  client_max_body_size 128m;
}

upstream api {
  server     127.0.0.1:8000  fail_timeout=10s;
}

upstream websockets {
  # The address below must match the one specified in puma's -b option.
  server     127.0.0.1:8100  fail_timeout=10s;
}

proxy_http_version 1.1;

# When Keep clients request a list of Keep services from the API server, the
# server will automatically return the list of available proxies if
# the request headers include X-External-Client: 1.  Following the example
# here, at the end of this section, add a line for each netmask that has
# direct access to Keep storage daemons to set this header value to 0.
geo $external_client {
  default        1;
  <span class="userinput">10.20.30.0/24</span>  0;
}

server {
  listen       <span class="userinput">[your public IP address]</span>:443 ssl;
  server_name  <span class="userinput">uuid_prefix.your.domain</span>;

  ssl on;
  ssl_certificate     <span class="userinput">/YOUR/PATH/TO/cert.pem</span>;
  ssl_certificate_key <span class="userinput">/YOUR/PATH/TO/cert.key</span>;

  index  index.html index.htm index.php;

  # Refer to the comment about this setting in the server section above.
  client_max_body_size 128m;

  location / {
    proxy_pass            http://api;
    proxy_redirect        off;
    proxy_connect_timeout 90s;
    proxy_read_timeout    300s;

    proxy_set_header      X-Forwarded-Proto https;
    proxy_set_header      Host $http_host;
    proxy_set_header      X-External-Client $external_client;
    proxy_set_header      X-Real-IP $remote_addr;
    proxy_set_header      X-Forwarded-For $proxy_add_x_forwarded_for;
  }
}

server {
  listen       <span class="userinput">[your public IP address]</span>:443 ssl;
  server_name  ws.<span class="userinput">uuid_prefix.your.domain</span>;

  ssl on;
  ssl_certificate     <span class="userinput">/YOUR/PATH/TO/cert.pem</span>;
  ssl_certificate_key <span class="userinput">/YOUR/PATH/TO/cert.key</span>;

  index  index.html index.htm index.php;

  location / {
    proxy_pass            http://websockets;
    proxy_redirect        off;
    proxy_connect_timeout 90s;
    proxy_read_timeout    300s;

    proxy_set_header      Upgrade $http_upgrade;
    proxy_set_header      Connection "upgrade";
    proxy_set_header      Host $host;
    proxy_set_header      X-Real-IP $remote_addr;
    proxy_set_header      X-Forwarded-For $proxy_add_x_forwarded_for;
  }
}
</code></pre>
</li>

<li><p>Restart Nginx:</p>

<pre><code>~$ <span class="userinput">sudo nginx -s reload</span>
</code></pre>

</li>

</ol>
</notextile>

h2. Prepare the API server deployment

{% assign railspkg = "arvados-api-server" %}
{% include 'install_rails_reconfigure' %}

{% include 'notebox_begin' %}
You can safely ignore the following messages if they appear while this command runs:
<pre>Don't run Bundler as root. Bundler can ask for sudo if it is needed, and installing your bundle as root will
break this application for all non-root users on this machine.</pre>
<pre>fatal: Not a git repository (or any of the parent directories): .git</pre>
{% include 'notebox_end' %}