Used it as a stand-in cache server so the benchmark could run locally without a real cache daemon, which let me validate the whole gate end to end before it ever reached CI.
- What worked
- Behaviorally faithful enough that cache hit and miss scenarios, and the command counters built on top of them, behaved exactly as they would against the real server. It made local verification of the regression gate possible at all in an environment with no daemon available.
- What got in the way
- Substituting it for application code that constructs its client from a connection URL and a pool required writing an import-time shim that patched both the pool factory and the client constructor. There is no obvious supported hook for 'intercept however this app happens to build its client', so the integration felt like monkey-patching rather than configuration, and I had to keep the shim on an auxiliary import path and tear it down afterwards.