There are many ways to deploy Alerta. It can be run as alertad
during development or testing but when run in a production environment,
it should always be deployed as a WSGI application. See the list
of real world examples below for different ways to run Alerta as
a WSGI application.
When deploying with Apache mod_wsgi, be aware that by default Apache
strips the Authentication header. This will cause you to receive
"Missing authorization API Key or Bearer Token" errors. This can be
fixed by setting WSGIPassAuthorization On
in the configuration
file for the site.
Running the Alerta API behind a web proxy can greatly simplify the Web UI setup which means you can completely avoid the potential for any cross-origin issues.
Also, if you run the API on an HTTPS/SSL endpoint then it can reduce the possibility of mixed content errors when a web application hosted on a HTTP endpoint tries to access resources on an HTTPS endpoint.
Example API configuration (extract)
This example nginx server is configured to serve the web UI from
the root /
path and reverse-proxy API requests to /api
to
the WSGI application running on port 8080:
server { listen 80 default_server deferred; access_log /dev/stdout main; location /api/ { proxy_pass http://backend/; proxy_set_header Host $host:$server_port; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; } location / { root /app; } } upstream backend { server localhost:8080 fail_timeout=0; }
The web UI configuration file :file:`config.js` for this setup would
simply be /api
for the endpoint
value, as follows:
'use strict'; angular.module('config', []) .constant('config', { 'endpoint' : "/api", 'provider' : "basic" });
The Alerta web UI is just a directory of static assets that can be served from any location. An easy and cheap way to serve the web UI is from an Amazon S3 bucket as a static website.
Note
Serving the Alerta web UI from a static web hosting site
will not work unless that domain is listed in the
CORS_ORIGINS
Alerta API server configuration settings.
Alerta supports several authentication mechanisms for both the API and the web UI and some key features of the web UI, like :ref:`watching alerts <watched alerts>`, are only available if authentication is enabled.
The API can be secured using :ref:`API keys` and the web UI can be secured using :ref:`Basic Auth <basic auth>` or an :ref:`OAuth <oauth2>` provider from either Google or Github.
If you plan to make the web UI accessible from a public URL it is strongly advised to :ref:`enforce authentication <Authentication>` and use HTTPS/SSL connections to the Alerta API to protect private alert data.
To restrict access to certain features use :ref:`roles <user roles>` and :ref:`customer views <customer views>`.
Alerta can scale horizontally, in the same way any other web application scales horizontally -- a load balancer handles the HTTP requests and distributes those requests between all available application servers.
Note
If using multiple API servers ensure the same SECRET_KEY
is used across all servers otherwise there will be problems
with web UI user logins.
To achieve high system availability the Alerta API should be deployed to scale out :ref:`horizontally <scalability>` and the MongoDB database should be deployed as a replica set.
There are some jobs that should be run periodically to keep the Alerta console clutter free. To timeout expired alerts and delete old closed alerts you need to trigger housekeeping.
This can be done with the alerta
command-line tool:
$ alerta housekeeping
This was not supported by earlier versions of the command-line tool
and cURL has to be used to access /management/housekeeping
.
The API key needs an admin scope if AUTH_REQUIRED is set to True.
It is suggested that you run housekeeping at regular intervals via
cron
. Every minute is a suitable interval.
By default, when you run housekeeping, Alerta will remove any alerts that have been expired or closed for 2 hours and any info messages that are 12 hours old. In some cases, these retention periods may be too long or too short for your needs. Bear in mind that Alerta is intended to reflect the here and now, so long deletion thresholds should be avoided. Where you do need to depart from the defaults, you can specify like this:
$ alerta housekeeping --expired 2 --info 12
In earlier versions of Alerta, a script called housekeepingAlerts.js was used for housekeeping. This is now deprecated.
:ref:`Heartbeats <heartbeats>` can be sent from any source to
ensure that a system is 'alive'. To generate alerts for stale
heartbeats the alerta
command-line tool can be used:
$ alerta heartbeats --alert
Again, this should be run at regular intervals via cron
or
some other scheduler.
Use the management endpoint :file:`/management/status` to keep
track of realtime statistics on the performance of the Alerta API
like alert counts and average processing time. For convenience,
these statistics can be viewed in the About page of the Alerta
web UI or using the alerta
command-line tool
:ref:`status <cli_status>` command.
Google analytics can be used to track usage of the Alerta web UI
console. Just create a new tracking code with the Google analytics
console and add it to the config.js
web console configuration
file:
'use strict'; angular.module('config', []) .constant('config', { 'endpoint' : "/api", 'provider' : "basic", 'tracking_id' : "UA-NNNNNN-N" // Google Analytics tracking ID });
Below are several different examples of how to run Alerta in production from a Debian vagrant box, an AWS EC2 instance, RedHat Openshift PaaS to a Docker container.
- Vagrant - deploy Alerta stand-alone or with Nagios, Zabbix, Riemann, Sensu or Kibana
- Heroku - deploy the Alerta API and the web ui to Heroku PaaS
- OpenShift - deploy the Alerta API to OpenShift Paas
- AWS EC2 - deploy Alerta to EC2 using AWS Cloudformation
- Docker - deploy Alerta to a docker container
- Packer - deploy Alerta to EC2 using Amazon AMIs
- Flask deploy - deploy Alerta as a generic Flask app
- Ansible - deploy Alerta using ansible on Centos 7