The drop(&mut sub) expression doesn't actually drop the subscriber value, only the mutable reference to it. This means the subscription will remain active since the actual subscriber isn't being dropped. To properly clean up resources and trigger unsubscription, this should be changed to drop(sub) without the reference operator.

Spotted by Diamond



Is this helpful? React 👍 or 👎 to let us know.

The spawned task in the request benchmark setup creates an infinite loop that continues running after the benchmark completes. This can lead to resource leakage as these background tasks accumulate across benchmark runs. Consider either:

1. Capturing the JoinHandle from tokio::spawn() and aborting it during teardown
2. Using a channel or atomic flag to signal the task to exit gracefully when the benchmark completes

This ensures proper cleanup of resources between benchmark iterations and prevents potential memory/thread leaks during extended benchmark runs.

Spotted by Diamond



Is this helpful? React 👍 or 👎 to let us know.

The 10ms timeout creates a potential race condition in the benchmark. Under high load or network latency, legitimate messages might arrive after the timeout expires, causing the benchmark to undercount messages and potentially report inaccurate results. Consider either:

1. Using a longer, more conservative timeout value (e.g., 100ms)
2. Implementing a more deterministic approach that doesn't rely on timing
3. Adding instrumentation to detect and report when messages might have been missed

This is particularly important for benchmarks that need to produce consistent, reproducible results across different environments.

Spotted by Diamond



Is this helpful? React 👍 or 👎 to let us know.