Implement setting to keep null values in events

[ycombinator]

Resolves #5522.

This PR proposes to add a new setting in Beats' configuration, keep_null, that will allow published events to contain null values in them. The default value of this setting will be false, which preserves the current behavior of Beats (i.e. prior to this setting being introduced).

Depending on the Beat, this setting can be specified at different places in the configuration.

Filebeat

Specify the setting under an input's configuration. For example:

filebeat.inputs:
- type: stdin
  enabled: true
  keep_null: true
  json.keys_under_root: false

With the above configuration, input text such as this:

{"test-null": null, "@timestamp":"2016-11-01T14:39:36.322", "type": 42,  "level":"INFO", "message":"test"}

Results in the following output event:

{"@timestamp":"2019-10-04T19:10:44.041Z","@metadata":{"beat":"filebeat","type":"_doc","version":"8.0.0"},"input":{"type":"stdin"},"ecs":{"version":"1.1.0"},"host":{"name":"Shaunaks-MacBook-Pro-Work.local"},"agent":{"hostname":"Shaunaks-MacBook-Pro-Work.local","id":"8ae6a937-ade6-482c-8f47-59e7dff17a44","version":"8.0.0","type":"filebeat","ephemeral_id":"def7a61b-ea7e-4825-97ba-87fb1257f3b7"},"log":{"offset":0,"file":{"path":""}},"json":{"test-null":null,"@timestamp":"2016-11-01T14:39:36.322","type":42,"level":"INFO","message":"test"}}

Notice, in particular, that the output event contains the json.test-null field whose value is null.

The setting can also be specified under Filebeat modules' input configuration.

Metricbeat

Specify the setting under a module's configuration. For example:

- module: http
  metricsets:
    - server
  host: "localhost"
  port: "8080"
  enabled: true
  keep_null: true

With the above configuration, the following HTTP request:

curl -H 'Content-Type: application/json' 'http://localhost:8080' -d '{ "foo": 17, "bar": null }'

Results in the following output event:

{"@timestamp":"2019-10-08T10:51:35.640Z","@metadata":{"beat":"metricbeat","type":"_doc","version":"8.0.0"},"host":{"hostname":"Shaunaks-MacBook-Pro-Work.local","name":"Shaunaks-MacBook-Pro-Work.local","architecture":"x86_64","os":{"kernel":"18.7.0","build":"18G103","platform":"darwin","version":"10.14.6","family":"darwin","name":"Mac OS X"},"id":"EF6274EA-462F-5316-A14A-850E7BFD8126"},"agent":{"id":"0b7c7d8b-d4b0-4d37-8704-a97231e6b460","version":"8.0.0","type":"metricbeat","ephemeral_id":"307c695c-ff2f-4ff1-812b-4e2b1eb56279","hostname":"Shaunaks-MacBook-Pro-Work.local"},"service":{"type":"http"},"http":{"server":{"foo":17,"bar":null}},"event":{"dataset":"http.server","module":"http"},"metricset":{"name":"server"},"ecs":{"version":"1.1.0"}}

Notice, in particular, that the output event contains the http.server.bar field whose value is null.

Auditbeat

Specify the setting under a module's configuration. For example:

- module: system
  datasets:
    - process
    - socket
  period: 1s
  keep_null: true

Packetbeat

Specify the setting under a protocol's configuration or the flows configuration. For example:

packetbeat.flows:
  timeout: 30s
  period: 10s
  keep_null: true

- type: http
  ports: [80, 8080, 8000, 5000, 8002]
  keep_null: true

Winlogbeat

Specify the setting under an event log's configuration. For example:

winlogbeat.event_logs:
  - name: Application
    keep_null: true

Heartbeat

Specify the setting under a monitor's configuration. For example:

heartbeat.monitors:
- type: http
  schedule: '@every 5s'
  urls: ["https://myhost:443"]
  keep_null: true

Functionbeat

Specify the setting under a function's configuration. For example:

functionbeat.provider.aws.endpoint: "s3.amazonaws.com"
functionbeat.provider.aws.deploy_bucket: "functionbeat-deploy"
functionbeat.provider.aws.functions:
  - name: cloudwatch
    enabled: true
    type: cloudwatch_logs
    description: "lambda function for cloudwatch logs"
    triggers:
      - log_group_name: /aws/lambda/my-lambda-function
    keep_null: true

[ycombinator]

Reviewers:

1. I'm not married to the name of this setting, emit_null_values. Please feel free to suggest alternatives 😄.

2. I would love some guidance on the right configuration level/scope for this setting. As of now it exists at the top-most/global level in a Beat's configuration. Is that what we want? Should be per-output? Per-input? Some kind of hierarchy with the setting's value at deeper levels overriding it's value from higher levels? Something else?

[urso]

1. I'm in favor for shorter but descriptive names like keep_null.

2. I'm no fan of global settings. Also in context of fleet every global setting is a pain. I'd say it is enough to have the setting per input only (keep change simple).

[ycombinator]

I'm no fan of global settings. Also in context of fleet every global setting is a pain. I'd say it is enough to have the setting per input only (keep change simple).

@urso If we make the setting per input then that implies it will only be available in Filebeat. However, other Beats could also generate events with null values, right?

[urso]

If we make the setting per input then that implies it will only be available in Filebeat. However, other Beats could also generate events with null values, right?

You would need to add these setting to metricbeat/heartbeat module setup as well. We already do this for 'processors', 'fields', 'tags' settings. Each beat passes these setting along when calling ConnectWith. ConnectWith then merges global and local settings into a list of event processors.

[ycombinator]

Hey @dedemorton, I was wondering if you could review this doc snippet before I copy it to other Beats' docs. Thanks!

[dedemorton]

LGTM.

[ycombinator]

@urso This PR is ready for review again. It has become rather large in size, primarily due to changes to the reference yml and docs files. If you want I can pull those changes out into a separate follow-up PR to make it easier to review this one. Just let me know!

[jsoriano]

I'd prefer not to add this to the reference config of all modules, we don't do it with other common options like fields, service.name in metricbeat or pipeline in filebeat.

[ycombinator]

Done in de6bb5d724212d844919b514480ab5cc2d732c32.

[urso]

KeepNull will be set be automatically set by the pipelineConnector in filebeat/channel/connector.go. This makes me wonder if we need to set Processing at all.

[ycombinator]

This makes me wonder if we need to set Processing at all.

I do think we need to set Processing here because of the DynamicFields config option. The pipelineConnector doesn't set it in its ConnectWith method.

makes sense.

[ycombinator]

@urso This is ready for another round of review, when you get a chance. Thanks!

[ycombinator]

jenkins, test this