Ayder Crash Sandbox — SIGKILL Durability Proof

2 min read Original article ↗

SIGKILL durability proof • per-visitor container • real disk

checking… sandbox group

Real SIGKILL. Real restart. Real committed offsets.

This page is a live durability demo. Each visitor gets a dedicated Ayder container with its own persisted /data volume. We produce events, kill the process with SIGKILL, restart the container, then prove offset semantics by committing and re-consuming with the same group.

port

healthy

commit semantics

Proof Summary

fast signal for HN readers: pass/fail + key numbers

No loss (found == expected)

Second consume (same group) is zero

restart no loss offset math same-group zero

Raw Proof JSON (exact API response)

Docker Forensics (inspect / events / logs tail)

What this proves

• Dedicated container per visitor (no interference) • Real SIGKILL via Docker • Manual restart (docker start) • Persistent volume stays mounted at /data Proof flow: 1) Produce N messages to topic "events" 2) SIGKILL the process (no graceful shutdown) 3) Restart container 4) Consume from offset=0 → count=N, next_offset=N 5) Commit via /broker/commit (commits last processed = next_offset-1) 6) Consume again (offset omitted → resume from committed+1) → 0 messages If you doubt: expand “Docker Forensics” and inspect kill/start/events/logs.

Tip: share the link — each person sees their own isolated sandbox.