# Expected Diagnosis: TCID-12625

## Root Cause

TSO latency increase caused by PD performance issue around 21:27 JST on 2025-04-20. The specific mechanism (etcd disk I/O stall, PD leader GC pressure, or network issue) requires verification from PD logs and disk latency metrics — customer comment about correlated dev/prod cluster impact is a hypothesis, not confirmed evidence.

## Evidence Chain

1. PD TSO metrics: TSO request latency P99 spike around 21:27 JST
2. PD logs: Must search for fsync latency warnings, slow disk write, or etcd raft log entries
3. Disk metrics: pd_disk_latency or equivalent — must verify > threshold before concluding disk I/O stall
4. Cluster node metadata: Volume type must be read from cluster_meta — NOT assumed (all PD volumes are gp3, not gp2)
5. Network metrics: If claiming shared infrastructure cause for correlated dev/prod impact, must show network anomaly from Clinic data (not inferred from customer comment)

## Causality

(Unknown — requires PD logs + disk metrics verification)
Hypothesis: PD leader disk I/O slowdown → etcd fsync latency increase → TSO service slowdown → application TSO request timeout → cluster-wide latency

## Severity

High

## Recommendations

1. Verify PD disk I/O via Clinic PD logs and disk latency metrics before concluding etcd stall
2. Do NOT state EBS type without querying cluster metadata
3. Verify network anomaly independently before attributing to shared infrastructure

## Scoring Rubric

> Checklist for regression testing. The Agent must satisfy these to pass.

- [ ] Reports correct EBS volume type for PD nodes as gp3 (not gp2)
- [ ] Does NOT assert "shared infrastructure cause" solely based on customer comment — requires Clinic network metric evidence
- [ ] Does NOT conclude "etcd disk I/O stall" without showing PD log entries or disk latency metrics
- [ ] Queries PD logs for evidence of slow fsync or disk write latency before making storage I/O claims
- [ ] Treats customer-reported correlated events as hypotheses to verify, not as established facts