Improve monitoring to measure the rate at which the postgres LockManager LWLock is acquired, rather than just tracking queue length during contention. This will allow for better visibility into system capacity and early warning of saturation, making it easier to assess how much headroom remains before user impact occurs. This may not be entirely possible - measuring this code path might be to expensive to instrument safely. But we should investigate the tradeoffs involved here. _This ticket was created from_ [_INC-8474_](https://app.incident.io/gitlab/incidents/8474) _using_ [_incident.io_](https://app.incident.io) 🔥