Feat/telemetry export clean

[theap06]

Closes #87

Summary

Adds a new telemetry exporter that periodically appends structured telemetry snapshots to a local file in JSON or CSV format for offline analysis.

What's New

• New sink: --sink=telemetry
• Output formats: JSON (NDJSON) or CSV
• Options:
  • file_path (required): Path to the output file
  • format (optional): json (default) or csv

Usage

# JSON (NDJSON, one object per line)
gcm nvml_monitor --sink=telemetry --sink-opt file_path=/var/log/gcm/telemetry.json --once

# CSV
gcm nvml_monitor --sink=telemetry --sink-opt file_path=/var/log/gcm/telemetry.csv --sink-opt format=csv --once

Example Output (JSON)

{"timestamp": "2026-03-04T21:31:22", "hostname": "node-42", "gpu_id": 3, "job_id": 91283, "job_user": "research_team", "gpu_util": 88, "mem_used_percent": 71, "temperature": 78, "power_draw": 310, "retired_pages_count_single_bit": 0, "retired_pages_count_double_bit": 0}

Implementation

• ~60 lines of code in gcm/exporters/telemetry.py
• Follows existing exporter conventions (@register, write(Log, SinkAdditionalParams))
• Works with gcm nvml_monitor, gcm slurm_monitor, and health check commands
• Auto-creates parent directories if needed

Testing

pytest gcm/tests/test_telemetry_exporter.py -v

[github-actions]

CI Commands

The following CI workflows run automatically on every push and pull request:

Workflow	What it runs
GPU Cluster Monitoring Python CI	lint, tests, typecheck, format, deb build, pyoxidizer builds
Go packages CI	shelper tests, format, lint

---

The following commands can be used by maintainers to trigger additional tests that require access to secrets:

Command	Description	Requires approval?
/metaci tests	Runs Meta internal integration tests (pytest)	Yes — a maintainer must trigger the command and approve the deployment request
/metaci integration tests	Same as above (alias)	Yes

Note: Only repository maintainers (OWNER association) can trigger /metaci commands. After commenting the command, a maintainer must also navigate to the Actions tab and approve the deployment to the graph-api-access environment before the jobs will run. See the approval guidelines for what to approve or reject.