feat: telemetry reporter for monitoring instances

Implements the reporter side of https://gitlab.com/postgres-ai/infra/-/work_items/52 — a small TS+Bun service that runs on each PostgresAI monitoring host and posts an hourly mini-healthcheck to the platform.

Summary

New top-level telemetry/ directory with:

• Collectors (each split into a pure parser + a thin runner so tests don't need root or docker):
  • oom: OOM kernel events from journalctl -k --since "<lookback>"
  • containers: faulty (exited / dead / restarting / unhealthy) docker containers via docker ps -a --format '{{json .}}'
  • memory: free RAM from /proc/meminfo MemAvailable, fallback to MemFree on very old kernels
  • disk: free bytes for the configured mount via fs.statfs
• Reporter: POSTs to <platform>/rpc/monitoring_instance_telemetry_report with the existing PostgresAI API token format
• Entry script: runs once on startup, then every PGAI_TELEMETRY_INTERVAL_SEC (default 3600s), with clean SIGTERM/SIGINT shutdown
• Dockerfile mounts /proc and a host disk path read-only

Configuration

env var	required	default
PGAI_PLATFORM_API_URL	yes	—
PGAI_API_TOKEN	yes	—
PGAI_MONITORING_INSTANCE_ID	yes	—
PGAI_TELEMETRY_DISK_PATH	no	/
PGAI_TELEMETRY_MEMINFO_PATH	no	/proc/meminfo
PGAI_TELEMETRY_OOM_LOOKBACK	no	24 hours ago
PGAI_TELEMETRY_INTERVAL_SEC	no	3600 (min 60)

Tests (TDD)

39 pass across 7 files, tsc --noEmit clean.

CI

telemetry:tests job in .gitlab-ci.yml, path-scoped to telemetry/**.

Companion

The platform-side hypertable, insert RPC, health evaluator, and alert dispatcher are in postgres-ai/platform-all!365.

Test plan

• bun test — 39/39 pass
• bun run typecheck — clean
• After companion MR merges and platform settings are configured: deploy one reporter container, observe at least one successful report in monitoring_instance_telemetry_latest
• Verify alert path: stop a container by hand and confirm Slack/email fires within the cron sweep window

Related: https://gitlab.com/postgres-ai/infra/-/work_items/52 (along with postgres-ai/platform-all!365)

🤖 Generated with Claude Code