system-config/doc/source/firehose_schema.rst
Monty Taylor e01ed4f066 Update some docs for opendev
There's a lot of these, so doing them in chunks. This fixes
the custom roles.

Remove the git and jjb docs, since we don't use them anymore.

Change-Id: I0c5b74f7b73315dac93bce6be0d920cddb94fb58
2019-04-20 09:41:45 -07:00

14 KiB

title

Firehose Schema

Firehose Schema

This attempts to document the topic and payload schema for all the services reporting to the firehose. However since much of what is reported to firehose is dynamically generated it is possible this document misses a case.

Gerrit

Messages on firehose for gerrit are generated using the germqtt project. For the most part these are basically identical to what gerrit returns on it's native event stream except over MQTT.

Topics

The topics for gerrit are generated dynamically. However, they follow a fairly straightforward pattern. The basic formula for this is:

gerrit/<git namespace>/<repo name>/<gerrit event>

So for example a typical topic would be:

gerrit/openstack/nova/comment-added

The git namespace and repo name are pretty self explanatory and are just from the git repository the change in gerrit is for. The event is defined in the gerrit event stream. You can see the full reference for topics in the Gerrit docs for Gerrit events. However, for simplicity the possible values are:

  • change-abandoned
  • change-merged
  • change-restored
  • comment-added
  • draft-published
  • hashtags-changed
  • merge-failed
  • patchset-created
  • ref-updated
  • reviewer-added
  • topic-changed

Payload

The payload for gerrit messages are basically the same JSON that gets returned by gerrit's event stream command. Instead of repeating the entire gerrit schema doc here just refer to gerrit's docs on the JSON payload which documents the contents of each JSON object and refer to the doc on Gerrit events for which JSON objects are included with which event type.

Launchpad

The messages sent to firehose for launchpad are generated using lpmqtt

Topics

The topics for lpmqtt follow a pretty simple formula:

launchpad/<project>/<event type>/<bug number>

the project is the launchpad project name, event type will always be "bug" (or not present). The intent of this was to be "bug" or "blueprint", but due to limitations in launchpad getting notifications from blueprints is not possible. The flexibility was left in the schema just in case this ever changes. The bug number is obviously the bug number from launchpad.

It's also worth noting that only the base topic is a guaranteed field. Depending on the notification email from launchpad some of the other fields may not be present. In those cases the topic will be populated left to right until a missing field is encountered.

Payload

The payload of messages is dynamically generated and dependent on the notification received from launchpad, and launchpad isn't always consistent in what fields are present in those notifications.

However, for bug event types there is a standard format. The fields which are always present for bugs (which should normally be the only message for firehose) are:

  • commenters
  • bug-reporter
  • bug-modifier
  • bug-number
  • event-type

The remaining fields are dynamic and depend on launchpad. An example message payload (with the body trimmed) for a bug is:

{
  "status": "Triaged",
  "project": "octavia",
  "assignee": "email@fakedomain.com",
  "bug-reporter": "Full name (username)",
  "event-type": "bug",
  "bug-number": "1680938",
  "commenters": ["username"]
  "tags": ["rfe"],
  "importance": "Medium",
  "bug-modifier": "Full Name (username)",
  "body": "notification body, often is just bug comment or summary",
}

Subunit Workers

The messages for the subunit workers are generated directly in the subunit gearman worker scripts.

Topics

The topics for the subunit workers follow a simple pattern:

gearman-subunit/<worker hostname>/<git namespace/<repo name>/<change number>

Where worker hostname is the host which processed the subunit file, as of right now there are 2, subunit-worker01 and subunit-worker02, but there may be more (or fewer) in the future. The git namespace and repo name are pretty self explanatory, and are just for the git repo under test that the subunit was emitted from. change number is the gerrit change number for the job that launched the tests the subunit is for.

Payload

The payload for the messages from the subunit workers is pretty straightforward json that contains 3 fields: status, build_uuid, and source_url.

An example is:

{
    'status': 'success',
    'build_uuid': '45f7c1ddbfd74c6aba94662623bd61b8'
    'source_url': 'A url',
}

Ansible

We have mqtt events emitted from ansible being run on puppet-master. These events are generated using a MQTT Ansible Callback Plugin.

Topics

The topics for ansible are a bit more involved than some of the other services publishing to firehose. It depends on the type of event that ansible just finished. There are 3 categories of events which have slightly different topic formulas (and payloads).

Playbook Events

Whenever a playbook action occurs the callback plugin will emit an event for it. The topics for playbook events fall into this pattern:

ansible/playbook/<playbook uuid>/action/<playbook action>/<status>

playbook uuid is pretty self explanatory here, it's the uuid ansible uses to uniquely identify the playbook being run. playbook action is the action that the event is for, this is either going to be start or finish. status is only set on finish and will be one of the following:

  • OK
  • FAILED

to indicate whether the playbook successfully executed or not.

Playbook Stats Events

At the end of a playbook these events are emitted for each host that tasks were run on. The topics for these events fall into the following pattern:

ansible/playbook/<playbook uuid>/stats/<hostname>

In this case playbook uuid is the same as above and the internal ansible unique playbook identifier. hostname here is the host that ansible was running tasks on as part of the playbook.

Task Events

At the end of each individual task the callback plugin will emit an event. Those events' topics fall into the following pattern:

ansible/playbook/<playbook uuid>/task/<hostname>/<status>

playbook uuid is the same as in the previous 2 event types. hostname is the hostname the task was executed on. status is the result of the task and will be one of the following:

  • OK
  • FAILED
  • UNREACHABLE

Payload

Just as with the topics the message payloads depend on the event type. Each event uses a JSON payload with slightly different fields.

Playbook Events

For playbook events the payload falls into this schema on playbook starts:

{
    "status": "OK",
    "host": <hostname>
    "session": <session id>,
    "playbook_name": <playbook name>,
    "playbook_id": <playbook uuid>,
    "ansible_type": "start",
}

When a playbook finishes the payload is slightly smaller and the schema is:

{
    "playbook_id": <playbook uuid>,
    "playbook_name": <playbook name>,
    "status": <status>,
}

In both cases playbook uuid is the same field from the topic. playbook name is the human readable name for the playbook. If one is set in the playbook this will be that. status will be whether the playbook was successfully executed or not. It will always be OK on starts (otherwise the event isn't emitted) but on failures, just like in the topic, this will be one of the following:

  • OK
  • FAILED

session id is a UUID generated by the callback plugin to uniquely identify the execution of the playbook. hostname is the hostname where the ansible playbook was launched. (which is not necessarily where tasks are being run)

An example of this from the system is for a start event:

{
    "status": "OK",
    "playbook_name": "localhost:!disabled",
    "ansible_type": "start",
    "host": "puppetmaster.openstack.org",
    "session": "14d6e568-2c75-11e7-bd24-bc764e048db9",
    "playbook_id": "5a95e9da-8d33-4dbb-a8b3-a77affc065d0"
}

and for a finish:

{
    "status": "FAILED",
    "playbook_name": "compute*.ic.openstack.org:!disabled",
    "playbook_id": "b259ac6d-6cb5-4403-bb8d-0ff2131c3d7a"
}

Playbook Stats Events

The schema for stats events is:

{
    "host": <hostname>,
    "ansible_host": <execute hostname>,
    "playbook_id": <playbook uuid>,
    "playbook_name": <playbook name>,
    "stats": {
        "unreachable": int,
        "skipped": int,
        "ok": int,
        "changed": int,
        "failures": int,
    }
}

playbook uuid is the same field from the topic. playbook name is the human readable name for the playbook. If one is set in the playbook this will be that. execute hostname is the hostname where the tasks were being executed, while hostname is the hostname where ansible launched the playbook. The stats subdict contains the task status counts where the key is the tasks statuses.

An example from the running system is:

{
    "playbook_name": "compute*.ic.openstack.org:!disabled",
    "host": "puppetmaster.openstack.org",
    "stats": {
        "unreachable": 0,
        "skipped": 5,
        "ok": 13,
        "changed": 1,
        "failures": 0
    },
    "playbook_id": "b259ac6d-6cb5-4403-bb8d-0ff2131c3d7a",
    "ansible_host": "controller00.vanilla.ic.openstack.org"
}

Task Events

The schema for tasks events is:

{
    "status": <status>,
    "host": <hostname>,
    "ansible_host": <execute hostname>,
    "session": <session id>,
    "ansible_type": "task",
    "playbook_name": <playbook name>,
    "playbook_id": <playbook uuid>,
    "ansible_task": <task name>,
    "ansible_result": <ansible result>
}

playbook uuid is the same field from the topic. playbook name is the human readable name for the playbook. If one is set in the playbook this will be that. execute hostname is the hostname where the tasks were being executed, while hostname is the hostname where ansible launched the playbook. task name, like the name implies, is the human readable name of the task executed. If one was specified in the playbook that will be the value. status is the result of the task and will be one of the following:

  • OK
  • FAILED
  • UNREACHABLE

session id is a UUID generated by the callback plugin to uniquely identify the execution of the playbook.

ansible result is a free form subdict that comes directly from ansible to completely describe the task that just finished. The structure here is fully dependent on ansible internals and the way that the task was invoked in the playbook. Note, that sometimes this can be quite large in size depending on the task and whether facts were enabled or not.

An example of a task event from the running system is:

{
    "status": "OK",
    "host": "puppetmaster.openstack.org",
    "session": "092aa3fa-2c73-11e7-bd24-bc764e048db9",
    "playbook_name": "compute*.ic.openstack.org:!disabled",
    "ansible_result": {
        "_ansible_parsed": true,
        "_ansible_no_log": false,
        "stdout": "",
        "changed": false,
        "stderr": "",
        "rc": 0,
        "invocation": {
            "module_name": "puppet",
            "module_args": {
                "logdest": "syslog",
                "execute": null,
                "facter_basename": "ansible",
                "tags": null,
                "puppetmaster": null,
                "show_diff": false,
                "certname": null,
                "manifest": "/opt/system-config/manifests/site.pp",
                "environment": "production",
                "debug": false,
                "noop": false,
                "timeout": "30m",
                "facts": null
            }
        },
        "stdout_lines": []
    },
    "ansible_type": "task",
    "ansible_task": "TASK: puppet : run puppet",
    "playbook_id": "b259ac6d-6cb5-4403-bb8d-0ff2131c3d7a",
    "ansible_host": "compute014.chocolate.ic.openstack.org"
}

Logstash Workers

The messages for the subunit workers are generated directly in the logstash gearman worker scripts.

Topics

The topics for the subunit workers follow a simple pattern:

gearman-logstash/<worker hostname>/<git namespace>/<repo name>/<change number>/<action>

Where worker hostname is the host which processed the log file. The git namespace and repo name are pretty self explanatory, and are just for the git repo under test that the log was generated. change number is the gerrit change number for the job that launched the tests the subunit is for. In the case of periodic or post queue jobs, this will either say periodic or post because there isn't an associated change number. The action field is the phase of the log processing that just completed. Right now the only possible value is retrieve_logs but there may be others in the future.

Payload

The payload for the messages from the logstash workers is pretty straightforward json that contains 3 fields: status, build_uuid, and source_url.

An example is:

{
    'status': 'success',
    'build_uuid': '45f7c1ddbfd74c6aba94662623bd61b8'
    'source_url': 'A url',
}