Introduction to Ansible Runner

Runner is intended to be most useful as part of automation and tooling that needs to invoke Ansible and consume its results. Most of the parameterization of the Ansible command line is also available on the Runner command line but Runner also can rely on an input interface that is mapped onto a directory structure, an example of which can be seen in the source tree.

Further sections in this document refer to the configuration and layout of that hierarchy. This isn’t the only way to interface with Runner itself. The Python module interface allows supplying these details as direct module parameters in many forms, and the command line interface allows supplying them directly as arguments, mimicking the behavior of ansible-playbook. Having the directory structure does allow gathering the inputs from elsewhere and preparing them for consumption by Runner, then the tooling can come along and inspect the results after the run.

This is best seen in the way Ansible AWX uses Runner where most of the content comes from the database (and other content-management components) but ultimately needs to be brought together in a single place when launching the Ansible task.

Runner Input Directory Hierarchy

This directory contains all necessary inputs. Here’s a view of the demo directory showing an active configuration.

Note that not everything is required. Defaults will be used or values will be omitted if they are not provided.

├── env
│   ├── envvars
│   ├── extravars
│   ├── passwords
│   ├── cmdline
│   ├── settings
│   └── ssh_key
├── inventory
│   └── hosts
└── project
    ├── test.yml
    └── roles
        └── testrole
            ├── defaults
            ├── handlers
            ├── meta
            ├── tasks
            ├── tests
            └── vars

The env directory

The env directory contains settings and sensitive files that inform certain aspects of the invocation of the Ansible process, an example of which can be found in the demo env directory. Each of these files can also be represented by a named pipe providing a bit of an extra layer of security. The formatting and expectation of these files differs slightly depending on what they are representing.



For an example see the demo envvars.

Ansible Runner will inherit the environment of the launching shell. This file (which can be in json or yaml format) represents the environment variables that will be added to the environment at run-time:

TESTVAR: exampleval



For an example see the demo extravars.

Ansible Runner gathers the extra vars provided here and supplies them to the Ansible Process itself. This file can be in either json or yaml format:

ansible_connection: local
test: val



For an example see the demo passwords.


We expect this interface to change/simplify in the future but will guarantee backwards compatibility. The goal is for the user of Runner to not have to worry about the format of certain prompts emitted from Ansible itself. In particular, vault passwords need to become more flexible.

Ansible itself is set up to emit passwords to certain prompts, these prompts can be requested (-k for example to prompt for the connection password). Likewise, prompts can be emitted via vars_prompt and also Ansible Vault.

In order for Runner to respond with the correct password, it needs to be able to match the prompt and provide the correct password. This is currently supported by providing a yaml or json formatted file with a regular expression and a value to emit, for example:

"^SSH password:\\s*?$": "some_password"
"^BECOME password.*:\\s*?$": "become_password"



Current Ansible Runner does not validate the command line arguments passed using this method so it is up to the playbook writer to provide a valid set of options. The command line options provided by this method are lower priority than the ones set by Ansible Runner. For instance, this will not override inventory or limit values.

Ansible Runner gathers command line options provided here as a string and supplies them to the Ansible Process itself. This file should contain the arguments to be added, for example:

--tags one,two --skip-tags three -u ansible --become



Currently only a single ssh key can be provided via this mechanism but this is set to change soon.

This file should contain the ssh private key used to connect to the host(s). Runner detects when a private key is provided and will wrap the call to Ansible in ssh-agent.

env/settings - Settings for Runner itself

The settings file is a little different than the other files provided in this section in that its contents are meant to control Runner directly.

  • idle_timeout: 600 If no output is detected from ansible in this number of seconds the execution will be terminated.

  • job_timeout: 3600 The maximum amount of time to allow the job to run for, exceeding this and the execution will be terminated.

  • pexpect_timeout: 10 Number of seconds for the internal pexpect command to wait to block on input before continuing

  • pexpect_use_poll: True Use poll() function for communication with child processes instead of select(). select() is used when the value is set to False. select() has a known limitation of using only up to 1024 file descriptors.

  • suppress_output_file: False Allow output from ansible to not be streamed to the stdout or stderr files inside of the artifacts directory.

  • suppress_ansible_output: False Allow output from ansible to not be printed to the screen.

  • fact_cache: 'fact_cache' The directory relative to artifacts where jsonfile fact caching will be stored. Defaults to fact_cache. This is ignored if fact_cache_type is different than jsonfile.

  • fact_cache_type: 'jsonfile' The type of fact cache to use. Defaults to jsonfile.

Process Isolation Settings for Runner

The process isolation settings are meant to control the process isolation feature of Runner.

  • process_isolation: False Enable limiting what directories on the filesystem the playbook run has access to.

  • process_isolation_executable: bwrap Path to the executable that will be used to provide filesystem isolation.

  • process_isolation_path: /tmp Path that an isolated playbook run will use for staging.

  • process_isolation_hide_paths: None Path or list of paths on the system that should be hidden from the playbook run.

  • process_isolation_show_paths: None Path or list of paths on the system that should be exposed to the playbook run.

  • process_isolation_ro_paths: None Path or list of paths on the system that should be exposed to the playbook run as read-only.

These settings instruct Runner to execute Ansible tasks inside a container environment. For information about building execution environments, see ansible-builder.

To execute Runner with an execution environment:

ansible-runner run --container-image my-execution-environment:latest --process-isolation -p playbook.yml .

See ansible-runner -h for other container-related options.


The Runner inventory location under the private data dir has the same expectations as inventory provided directly to ansible itself. It can be either a single file or script or a directory containing static inventory files or scripts. This inventory is automatically loaded and provided to Ansible when invoked and can be further overridden on the command line or via the ANSIBLE_INVENTORY environment variable to specify the hosts directly. Giving an absolute path for the inventory location is best practice, because relative paths are interpreted relative to the current working directory which defaults to the project directory.


The Runner project directory is the playbook root containing playbooks and roles that those playbooks can consume directly. This is also the directory that will be set as the current working directory when launching the Ansible process.


Runner has the ability to execute modules directly using Ansible ad-hoc mode.


Runner has the ability to execute Roles directly without first needing a playbook to reference them. This directory holds roles used for that. Behind the scenes, Runner will generate a playbook and invoke the Role.

Runner Artifacts Directory Hierarchy

This directory will contain the results of Runner invocation grouped under an identifier directory. This identifier can be supplied to Runner directly and if not given, an identifier will be generated as a UUID. This is how the directory structure looks from the top level:

├── artifacts
│   └── identifier
├── env
├── inventory
├── profiling_data
├── project
└── roles

The artifact directory itself contains a particular structure that provides a lot of extra detail from a running or previously-run invocation of Ansible/Runner:

├── artifacts
│   └── 37f639a3-1f4f-4acb-abee-ea1898013a25
│       ├── fact_cache
│       │   └── localhost
│       ├── job_events
│       │   ├── 1-34437b34-addd-45ae-819a-4d8c9711e191.json
│       │   ├── 2-8c164553-8573-b1e0-76e1-000000000006.json
│       │   ├── 3-8c164553-8573-b1e0-76e1-00000000000d.json
│       │   ├── 4-f16be0cd-99e1-4568-a599-546ab80b2799.json
│       │   ├── 5-8c164553-8573-b1e0-76e1-000000000008.json
│       │   ├── 6-981fd563-ec25-45cb-84f6-e9dc4e6449cb.json
│       │   └── 7-01c7090a-e202-4fb4-9ac7-079965729c86.json
│       ├── rc
│       ├── status
│       └── stdout

The rc file contains the actual return code from the Ansible process.

The status file contains one of three statuses suitable for displaying:

The stdout file contains the actual stdout as it appears at that moment.

Runner Artifact Job Events (Host and Playbook Events)

Runner gathers the individual task and playbook events that are emitted as part of the Ansible run. This is extremely helpful if you don’t want to process or read the stdout returned from Ansible as it contains much more detail and status than just the plain stdout. It does some of the heavy lifting of assigning order to the events and stores them in json format under the job_events artifact directory. It also takes it a step further than normal Ansible callback plugins in that it will store the stdout associated with the event alongside the raw event data (along with stdout line numbers). It also generates dummy events for stdout that didn’t have corresponding host event data:

  "uuid": "8c164553-8573-b1e0-76e1-000000000008",
  "parent_uuid": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
  "counter": 5,
  "stdout": "\r\nTASK [debug] *******************************************************************",
  "start_line": 5,
  "end_line": 7,
  "event": "playbook_on_task_start",
  "event_data": {
    "playbook": "test.yml",
    "playbook_uuid": "34437b34-addd-45ae-819a-4d8c9711e191",
    "play": "all",
    "play_uuid": "8c164553-8573-b1e0-76e1-000000000006",
    "play_pattern": "all",
    "task": "debug",
    "task_uuid": "8c164553-8573-b1e0-76e1-000000000008",
    "task_action": "debug",
    "task_path": "\/home\/mjones\/ansible\/ansible-runner\/demo\/project\/test.yml:3",
    "task_args": "msg=Test!",
    "name": "debug",
    "is_conditional": false,
    "pid": 10640
  "pid": 10640,
  "created": "2018-06-07T14:54:58.410605"

If the playbook runs to completion without getting killed, the last event will always be the stats event:

  "uuid": "01c7090a-e202-4fb4-9ac7-079965729c86",
  "counter": 7,
  "stdout": "\r\nPLAY RECAP *********************************************************************\r\n\u001b[0;32mlocalhost,\u001b[0m                 : \u001b[0;32mok=2   \u001b[0m changed=0    unreachable=0    failed=0   \r\n",
  "start_line": 10,
  "end_line": 14,
  "event": "playbook_on_stats",
  "event_data": {
    "playbook": "test.yml",
    "playbook_uuid": "34437b34-addd-45ae-819a-4d8c9711e191",
    "changed": {

    "dark": {

    "failures": {

    "ok": {
      "localhost,": 2
    "processed": {
      "localhost,": 1
    "skipped": {

    "artifact_data": {

    "pid": 10640
  "pid": 10640,
  "created": "2018-06-07T14:54:58.424603"


The Runner module interface presents a programmatic interface to these events that allow getting the final status and performing host filtering of task events.

Runner Profiling Data Directory

If resource profiling is enabled for Runner the profiling_data directory will be populated with a set of files containing the profiling data:

├── profiling_data
│   ├── 0-34437b34-addd-45ae-819a-4d8c9711e191-cpu.json
│   ├── 0-34437b34-addd-45ae-819a-4d8c9711e191-memory.json
│   ├── 0-34437b34-addd-45ae-819a-4d8c9711e191-pids.json
│   ├── 1-8c164553-8573-b1e0-76e1-000000000006-cpu.json
│   ├── 1-8c164553-8573-b1e0-76e1-000000000006-memory.json
│   └── 1-8c164553-8573-b1e0-76e1-000000000006-pids.json

Each file is in JSON text format. Each line of the file will begin with a record separator (RS), continue with a JSON dictionary, and conclude with a line feed (LF) character. The following provides an example of what the resource files may look like. Note that that since the RS and LF are control characters, they are not actually printed below:

==> 0-525400c9-c704-29a6-4107-00000000000c-cpu.json <==
{"timestamp": 1568977988.6844425, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 97.12799768097156}
{"timestamp": 1568977988.9394386, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 94.17538298892688}
{"timestamp": 1568977989.1901696, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 64.38272588006255}
{"timestamp": 1568977989.4594045, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 83.77387744259856}

==> 0-525400c9-c704-29a6-4107-00000000000c-memory.json <==
{"timestamp": 1568977988.4281094, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 36.21484375}
{"timestamp": 1568977988.6842303, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 57.87109375}
{"timestamp": 1568977988.939303, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 66.60546875}
{"timestamp": 1568977989.1900482, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 71.4609375}
{"timestamp": 1568977989.4592078, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 38.25390625}

==> 0-525400c9-c704-29a6-4107-00000000000c-pids.json <==
{"timestamp": 1568977988.4284189, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 5}
{"timestamp": 1568977988.6845856, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 6}
{"timestamp": 1568977988.939547, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 8}
{"timestamp": 1568977989.1902773, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 13}
{"timestamp": 1568977989.4593227, "task_name": "Gathering Facts", "task_uuid": "525400c9-c704-29a6-4107-00000000000c", "value": 6}
  • Resource profiling data is grouped by playbook task.

  • For each task, there will be three files, corresponding to cpu, memory and pid count data.

  • Each file contains a set of data points collected over the course of a playbook task.

  • If a task executes quickly and the polling rate for a given metric is large enough, it is possible that no profiling data may be collected during the task’s execution. If this is the case, no data file will be created.