Cloud Logging

Cloud Logging

This skill provides utilities for analyzing logs, errors, and system health across Google Cloud deployments.

• Prioritize Logs with severity >= "ERROR" as they tend to be less verbose.
• Constrain logs around the time of investigation (possibly going back a few hours/days to find smoking guns).

Logs DOs and DONT's

Ingesting Logs into an LMM memory can easily saturate your context.

• [BAD] DO NOT ingest big files. Antipattern: ReadFile path/to/access_logs_raw.json.
  • It's ok to get raw logs, but also use jq or scripts to read them without polluting your context window.
  • [GOOD] Use simple maniuplation like: jq -r '.[] | .httpRequest.status' to get a list of statuses.
  • [GOOD] Reduce size with scripts such as scripts/cloudlogging2apachelogs.py <big_logfile.json> (provided in the skill)
• [BAD] Do not call long-running calls without precautions, eg gcloud logging read
  • [GOOD] Rather prepend some reasonable timeout 60 gcloud logging read ..., or
  • [BEST] Use the run_shell_command tool with is_background: true and poll the results being dumped to file periodically.

Bundled Scripts

cloudlogging2apachelogs.py

Converts GCP Cloud Logging JSON exports into a format loosely resembling Apache Combined Log format. This is much more token-efficient for the LLM.

Usage:

python3 scripts/cloudlogging2apachelogs.py path/to/logs.json

Testing: You can verify the script works by running its test:

python3 scripts/cloudlogging2apachelogs_test.py

Bundled Assets

• assets/sample_logs.json: A small sample of GCP Cloud Logging JSON for testing conversion scripts.