Recipe: Tail Log Files

Use file_tail input when events are written to a local file as NDJSON.

Tail an application log

blazerules_agent \
  --rules rules.yaml \
  --input file_tail \
  --path /var/log/app/events.ndjson \
  --batch-size 2048 \
  --flush-ms 250 \
  --output ndjson \
  --output-path /var/log/blazerules/decisions.ndjson \
  --service checkout \
  --source file_tail

Python equivalent: start the same tailing agent

import subprocess

subprocess.Popen([
    "blazerules_agent",
    "--rules", "rules.yaml",
    "--input", "file_tail",
    "--path", "/var/log/app/events.ndjson",
    "--batch-size", "2048",
    "--flush-ms", "250",
    "--output", "ndjson",
    "--output-path", "/var/log/blazerules/decisions.ndjson",
    "--service", "checkout",
    "--source", "file_tail",
])

The agent starts at the end of the file and evaluates new lines as they arrive.

Input Example

{"event_id":"e1","level":"info","service":"checkout","message":"payment started","amount":30}
{"event_id":"e2","level":"error","service":"checkout","message":"payment failed","amount":900}

Rules can target any fields in the line:

ruleset:
  rules:
    - id: checkout_error
      action: REVIEW
      conditions:
        and:
          - field: level
            op: eq
            value: error
          - field: message
            op: contains
            value: payment

Rotation Behavior

file_tail is intentionally simple: it follows appended lines on one path. If your log system rotates files aggressively, point the agent at the stable symlink or use Kubernetes/container log paths that remain stable for the container lifetime.