A simple tool for tracking the output of automated ansible job runs using JSON files.

With ansible running periodically and in automated runs some form of tracking is needed to identify when playbooks fail and a particular system needs attention. One nice thing about ansible is that it scales down well for simple cases where you just want to manage a few machines. However, ansible AWX can be somewhat overkill for simple cases and the focus seems to be on virtual server deployments rather than, for example, physical desktops.

ansible-json-monitor involves just a simple command-line tool - ajmon - which takes JSON output files from ansible runs and produces a condensed summary. There is no always-on service or REST APIs requiring a master node. Just a directory full of files.

To get JSON output from ansible you need to set ANSIBLE_STDOUT_CALLBACK=json in the environment or ansible.cfg file to select the callback plugin for JSON output. With recent releases of ansible this plugin is distributed in the ansible.posix collection. As this goes to standard output, you need shell redirection to save it to a file which should then be compressed. To try this out we'll need something along these lines:

ANSIBLE_STDOUT_CALLBACK=json ansible-playbook plays.yml > output.json
gzip output.json

We now have a file named output.json.gz. We can get a basic overview of the results:

ajmon -f output.json.gz
CHANGED  FAILURES    OK  SKIPPED        DATE  HOST
      0         1   163       37  2019-06-07  localhost

By passing a hostname as the first normal argument to ajmon, we get a numbered list of all the tasks along with an indication of any that failed or which involved a change. Furthermore, the task number can be specified to get more details on the run of a specific task. The output in that case varies depending on the action.

There's also options to filter and sort tables by particular fields so for example, it is simple to limit the summary to those involving failures.

Ansible typically starts by running the setup action to gather a whole range of "facts" about the target system. ajmon can pick out these facts and present a number of different tables using based on the extracted data. So, for example, you can see the operating system versions as follows:

ajmon -f output.json.gz os
HOST    SYSTEM  DISTRIBUTION VERSION KERNEL
example FreeBSD FreeBSD         12.0 12.0-RELEASE-p8

For full details of the available options, there is a man page included.

To be useful, the tool needs a directory full of compressed json files containing the output of ansible runs.

At the moment, this primarily assumes one JSON file per-host which maps to a typical ansible-pull style setup. Given machines all sharing a file server, a shared checkout of the playbooks in a particular directory also works well. The following is an example shell script for running ansible, though it may require tweaks such as to use ansible-pull.

#!/bin/sh
set -e
exec >/dev/null 2>&1
log="/path/to/logs/staging/$HOST.json.gz"
cd /path/to/playbooks
ANSIBLE_STDOUT_CALLBACK=json ANSIBLE_RETRY_FILES_ENABLED=0 ansible-playbook -l $HOSTS -c local site.yml | su user -c "gzip -c > $log && mv $log /path/to/logs/ansible/."

This might typically be installed as /etc/cron.daily/ansible. If you're managing desktops, it can also be useful to arrange for ansible to run on boot.

Using su may not be necessary depending on how you have permissions configured. Sending the output to a staging directory and moving it over at the end avoids issues with potentially incomplete files. The ANSIBLE_RETRY_FILES_ENABLED option is required because ansible notifies about retry files to stdout resulting in an invalid JSON file.

This only keeps one output file per-host. It would certainly be possible to keep old ones but the single file avoids the need to cleanup old files and old files aren't especially useful anyway.