Vault barrier lock latency during concurrent secret requests

When HashiCorp Vault utilizes the Raft storage backend, performance degradation is sometimes observed during high-concurrency bursts targeting the same secret path. This behavior appears most prominently during 'thundering herd' scenarios where thousands of clients attempt to access or renew leases simultaneously.

The internal barrier mechanism is designed to ensure atomicity across storage operations. However, while linear scaling is expected with additional nodes, the specific overhead introduced by lock contention during massive atomic write operations or high-frequency lease renewals remains poorly defined relative to storage backend saturation.

How does the Vault barrier lock impact P99 latency when multiple concurrent requests compete for a single secret path? Is there a documented threshold where Raft storage contention becomes the primary bottleneck compared to cryptographic processing limits?