Ops Agent fails intermittently to execute type modify_fields instruction

[PawelKGalka]

Describe the bug
Havin exemplary field in jsonPayload.auditLog.rawData.txTimeMillis it is not always parsed to integer in Logs Explorer and remains as string

To Reproduce
Steps to reproduce the behavior:

1. Start a GCE VM with image RHEL 8
2. Install Ops Agent version 2.46.1
3. Customize the agent with configuration

logging:
  receivers:
    messages:
      type: files
      include_paths:
      - /var/log/messages
    secure:
      type: files
      include_paths:
      - /var/log/secure
    applogs_tcp:
      type: tcp
      format: json
      listen_port: 5160
  processors:
    parse:
      type: parse_json
      time_key: jsonPayload.timestamp
      time_format: "%Y-%m-%dT%H:%M:%S.%L%Z"
    adjust_fields_to_stackdriver:
      type: modify_fields
      fields:
        jsonPayload.auditLog.rawData.txTimeMillis:
          type: integer
  service:
    pipelines:
      default_pipeline:
        receivers: [messages, secure]
      applogs_pipeline:
        receivers: [applogs_tcp]
        processors: [parse, adjust_fields_to_stackdriver]

and restart the agent
5. See some values are still in String format

Expected behavior
Value is in integer format always

Environment (please complete the following information):

• Project ID - can send via email
• VM ID - can send via email
• VM distro / OS: RHEL8
• Ops Agent version 2.46.1
• Ops Agent configuration
• Ops Agent log

Additional context
NA

[jefferbrecht]

Hi @PawelKGalka, we recently released Ops Agent 2.47.0 which upgrades the version of Fluent Bit used internally. Are you able to upgrade to the latest version and reproduce the issue still?

If you're still able to reproduce this, can you provide any sample input logs that you're sending to that TCP port so that we can try to reproduce this ourselves? Feel free to open a Cloud Support case as well which would let us troubleshoot much more closely (you can reference this GitHub issue in that case if you'd like).