django-watchman

django-watchman exposes a status endpoint for your backing services like databases, caches, etc.

Documentation

The full documentation is at http://django-watchman.rtfd.org.

Testimonials

We're in love with django-watchman. External monitoring is a vital part of our service offering. Using django-watchman we can introspect the infrastructure of an application via a secure URL. It's very well written and easy to extend. We've recommended it to many of our clients already.

— Hany Fahim, CEO, VM Farms.

Quickstart

1. Install django-watchman:

   pip install django-watchman
   

2. Add watchman to your INSTALLED_APPS setting like this:

   INSTALLED_APPS = (
       ...
       'watchman',
   )
   

3. Include the watchman URLconf in your project urls.py like this:

   re_path(r'^watchman/', include('watchman.urls')),
   

4. Start the development server and visit http://127.0.0.1:8000/watchman/ to get a JSON response of your backing service statuses:

   {
       "databases": [
           {
               "default": {
                   "ok": true
               }
           }
       ],
       "caches": [
           {
               "default": {
                   "ok": true
               }
           }
       ],
       "storage": {"ok": true}
   }
   

Pycon Canada Presentation (10 minutes)

Features

Human-friendly dashboard

Visit http://127.0.0.1:8000/watchman/dashboard/ to get a human-friendly HTML representation of all of your watchman checks.

Token based authentication

If you want to protect the status endpoint, you can use the WATCHMAN_TOKENS setting. This is a comma-separated list of tokens. When this setting is added, you must pass one of the tokens in as the watchman-token GET parameter:

GET http://127.0.0.1:8000/watchman/?watchman-token=:token

Or by setting the Authorization: WATCHMAN-TOKEN header on the request:

curl -X GET -H "Authorization: WATCHMAN-TOKEN Token=\":token\"" http://127.0.0.1:8000/watchman/

If you want to change the token name, you can set the WATCHMAN_TOKEN_NAME. The value of this setting will be the GET parameter that you must pass in:

WATCHMAN_TOKEN_NAME = 'custom-token-name'

GET http://127.0.0.1:8000/watchman/?custom-token-name=:token

DEPRECATION WARNING: WATCHMAN_TOKEN was replaced by the WATCHMAN_TOKENS setting to support multiple authentication tokens in django-watchman 0.11. It will continue to work until it's removed in django-watchman 1.0.

Custom authentication/authorization

If you want to protect the status endpoint with a customized authentication/authorization decorator, you can add WATCHMAN_AUTH_DECORATOR to your settings. This needs to be a dotted-path to a decorator, and defaults to watchman.decorators.token_required:

WATCHMAN_AUTH_DECORATOR = 'django.contrib.admin.views.decorators.staff_member_required'

Note that the token_required decorator does not protect a view unless WATCHMAN_TOKENS is set in settings.

Custom checks

django-watchman allows you to customize the checks which are run by modifying the WATCHMAN_CHECKS setting. In settings.py:

WATCHMAN_CHECKS = (
    'module.path.to.callable',
    'another.module.path.to.callable',
)

You can also import the watchman.constants to include the DEFAULT_CHECKS and PAID_CHECKS in your settings.py:

from watchman import constants as watchman_constants

WATCHMAN_CHECKS = watchman_constants.DEFAULT_CHECKS + ('module.path.to.callable', )

Checks take no arguments, and must return a dict whose keys are applied to the JSON response.

Use the watchman.decorators.check decorator to capture exceptions:

from watchman.decorators import check

def custom_check():
    return {"custom_check": _custom_check()}

@check
def _custom_check():
    return {"ok": True, "extra_info": "if helpful"}

In the absence of any checks, a 404 is thrown, which is then handled by the json_view decorator.

Run a subset of available checks

A subset of checks may be run, by passing ?check=module.path.to.callable&check=... in the request URL. Only the callables given in the querystring which are also in WATCHMAN_CHECKS should be run, eg:

curl -XGET http://127.0.0.1:8080/watchman/?check=watchman.checks.caches

Skip specific checks

You can skip any number of checks, by passing ?skip=module.path.to.callable&skip=... in the request URL. Only the checks in WATCHMAN_CHECKS which are not in the querystring should be run, eg:

curl -XGET http://127.0.0.1:8080/watchman/?skip=watchman.checks.email

Check a subset of databases or caches

If your application has a large number of databases or caches configured, watchman may open too many connections as it checks each database or cache.

You can set the WATCHMAN_DATABASES or WATCHMAN_CACHES settings in order to override the default set of databases and caches to be monitored.

Ping

If you want to simply check that your application is running and able to handle requests, you can call ping:

GET http://127.0.0.1:8000/watchman/ping/

It will return the text pong with a 200 status code. Calling this doesn't run any of the checks.

Bare status view

If you would like a "bare" status view (one that doesn't report any details, just HTTP 200 if checks pass, and HTTP 500 if any checks fail), you can use the bare_status view by putting the following into urls.py:

import watchman.views
# ...
re_path(r'^status/?$', watchman.views.bare_status),

Django management command

You can also run your checks without starting the webserver and making requests. This can be useful for testing your configuration before enabling a server, checking configuration on worker servers, etc. Run the management command like so:

python manage.py watchman

By default, successful checks will not print any output. If all checks pass successfully, the exit code will be 0. If a check fails, the exit code will be 1, and the error message including stack trace will be printed to stderr.

If you'd like to see output for successful checks as well, set verbosity to 2 or higher:

python manage.py watchman -v 2
{"storage": {"ok": true}}
{"caches": [{"default": {"ok": true}}]}
{"databases": [{"default": {"ok": true}}]}

If you'd like to run a subset of checks, use -c and a comma-separated list of python module paths:

python manage.py watchman -c watchman.checks.caches,watchman.checks.databases -v 2
{"caches": [{"default": {"ok": true}}]}
{"databases": [{"default": {"ok": true}}]}

If you'd like to skip certain checks, use -s and a comma-separated list of python module paths:

python manage.py watchman -s watchman.checks.caches,watchman.checks.databases -v 2
{"storage": {"ok": true}}

Use -h to see a full list of options:

python manage.py watchman -h

X-Watchman-Version response header

Watchman can return the version of watchman which is running to help you keep track of whether or not your sites are using an up-to-date version. This is disabled by default to prevent any unintended information leakage for websites without authentication. To enable, update the EXPOSE_WATCHMAN_VERSION setting:

EXPOSE_WATCHMAN_VERSION = True

Custom response code

By default, watchman will return a 500 HTTP response code, even if there's a failing check. You can specify a different response code for failing checks using the WATCHMAN_ERROR_CODE setting:

WATCHMAN_ERROR_CODE = 200

Logging

watchman includes log messages using a logger called watchman. You can configure this by configuring the LOGGING section of your Django settings file.

Here is a simple example that would log to the console:

LOGGING = {
    'version': 1,
    'disable_existing_loggers': False,
    'handlers': {
        'console': {
            'class': 'logging.StreamHandler',
        },
    },
    'loggers': {
        'watchman': {
            'handlers': ['console'],
            'level': 'DEBUG',
        },
    },
}

More information is available in the Django documentation.

APM (Datadog, New Relic)

If you're using APM and watchman is being often hit for health checks (such as an ELB on AWS), you will find some stats based on averages will be affected (average transaction time, apdex, etc):

You can disable APM instrumentation for watchman by using the WATCHMAN_DISABLE_APM setting:

WATCHMAN_DISABLE_APM = True

This currently supports the following agents:

• Datadog
• New Relic

Please open an issue if there's another APM you use which is being affected.

Available checks

caches

For each cache in django.conf.settings.CACHES:

• Set a test cache item
• Get test item
• Delete test item

databases

For each database in django.conf.settings.DATABASES:

• Verify connection by calling connections[database].introspection.table_names()

email

Send a test email to [email protected] using django.core.mail.send_mail.

If you're using a 3rd party mail provider, this check could end up costing you money, depending how aggressive you are with your monitoring. For this reason, this check is not enabled by default.

For reference, if you were using Mandrill, and hitting your watchman endpoint once per minute, this would cost you ~$5.60/month.

Custom Settings

• WATCHMAN_EMAIL_SENDER (default: [email protected]): Specify an email to be the sender of the test email
• WATCHMAN_EMAIL_RECIPIENTS (default: [[email protected]]): Specify a list of email addresses to send the test email
• WATCHMAN_EMAIL_HEADERS (default: {}): Specify a dict of custom headers to be added to the test email

storage

Using django.core.files.storage.default_storage:

• Write a test file
• Check the test file's size
• Read the test file's contents
• Delete the test file

By default the test file gets written on the root of the django MEDIA_ROOT. If for whatever reasons this path is not writable by the user that runs the application you can override it by setting WATCHMAN_STORAGE_PATH to a specific path. Remember that this must be within the MEDIA_ROOT, which by default is your project root. In settings.py:

WATCHMAN_STORAGE_PATH = "/path_to_your_app/foo/bar/"

If the MEDIA_ROOT is already defined:

from os.path import join as joinpath
WATCHMAN_STORAGE_PATH = joinpath(MEDIA_ROOT, "foo/bar")

Default checks

By default, django-watchman will run checks against your databases (watchman.checks.databases), caches (watchman.checks.caches), and storage (watchman.checks.storage).

Paid checks

Paid checks are checks which may cost you money if they are run regularly.

Currently there is only one "paid" check - watchman.checks.email. Many times email is sent using managed services like SendGrid or Mailgun. You can enable it by setting the WATCHMAN_ENABLE_PAID_CHECKS to True, or by overriding the WATCHMAN_CHECKS setting.

Trying it out with Docker

A sample project is available along with a Dockerfile to make it easy to try out django-watchman. It includes examples of how to write simple custom checks.

One of the custom checks will always fail, so if you want to see what responses look like with 100% succeeding checks, add ?skip=sample_project.checks.fail_custom_check

Requirements

• Docker <https://www.docker.com/get-docker>

Instructions

1. Build and run the Docker image with the current local code: make run
2. Visit watchman json endpoint in your browser: http://127.0.0.1:8000/watchman/
3. Visit watchman json endpoint in your browser: http://127.0.0.1:8000/watchman/?skip=sample_project.checks.fail_custom_check
4. Visit watchman dashboard in your browser: http://127.0.0.1:8000/watchman/dashboard/
5. Visit watchman dashboard in your browser: http://127.0.0.1:8000/watchman/dashboard/?skip=sample_project.checks.fail_custom_check
6. Visit watchman ping in your browser: http://127.0.0.1:8000/watchman/ping/
7. Visit watchman bare status in your browser: http://127.0.0.1:8000/watchman/bare/