# Expected Diagnosis: system-unavailability-during-incident-window

This case should converge on an infrastructure-capacity failure path rather than
stopping at a query-only explanation.

The expected answer should make these points explicit:

- Multiple coprocessor workers became unavailable in a short interval, causing a
  sharp capacity drop before deeper query-level symptoms appeared.
- The single remaining worker became overloaded and entered repeated restart or
  memory-failure behavior, which amplified the incident into near-outage
  conditions.
- The causal chain should connect worker loss, request backlog, coprocessor or
  gRPC latency explosion, TiDB query timeout, and QPS collapse.
- Recommendations should focus on replica floor, autoscaling guardrails, worker
  health monitoring, and follow-up on the initial disappearance trigger.