When Riffle detects disk corruption, it marks the server as unhealthy. However, the await-tree log still shows hung requests reading from the broken disk. In fact, we already have an I/O timeout layer: if a request’s duration exceeds the configured threshold, it should be stopped.
