waits-latches-spinlocks

Waits, latches, and spinlocks

Procedure

1. Use the collection map to choose the scenario. Read its Markdown entry point and retained readme.txt instructions.
2. Confirm platform, SQL build/tier, instance, SQL tools, and the scenario's WinDbg/symbol or stress-driver dependencies. Review startup flags, authentication, local paths, and resource limits.
3. Use a disposable lab. Obtain approval before restarts, debugger attachment, Resource Governor changes, heavy traffic, bulk inserts, or configuration changes. Supply credentials privately; do not reuse literal demo passwords.
4. Capture baseline requests, wait/latch/spinlock counters, and CPU/log metrics. Run the approved workload in the intended sessions, then compare deltas and stacks rather than equating a cumulative wait count with the root cause.
5. Resolve spinlock type IDs and internal symbols against the current build. Captured driver output is reference material, not proof of current behavior.
6. Stop the specific lab processes, close transactions, detach the debugger, and follow the chosen cleanup while restoring prior settings.

Do not run contention or failure reproductions in production. Follow secret checks before publishing logs or scripts.