Start with a failure you can explain
This repository isolates a case in the MongoDB Rust driver where polling a change stream after its collection is dropped can enter a busy loop. The reproduction also demonstrates why a runtime watchdog can fail to get scheduled when the task stops yielding.
Reduce the problem, then report it
The investigation narrows the behavior to a small runnable example and traces a regression through the driver’s cursor changes. That makes the failure easier for maintainers to reproduce and evaluate.
The upstream report is RUST-2465. The repository records the tested driver versions and reproduction steps; it is a record of the investigation, rather than a claim about every current driver release.
The work behind reliable systems
Performance and reliability work often begins with evidence: a trace, a repeatable failure, and a hypothesis that survives a smaller test. This is one public example of that process.