mlx Six Loops, No Timeout: The MLX Bug That Hangs a Mac Cluster Forever
MLX issue #3910: when a peer dies mid-collective, JACCL spins forever with zero errno, zero exception, zero log output. I read the source. It is six unbounded loops, not one, plus a second unreported bug with a negative-offset pointer. Here is what is actually broken and why Rust would not have save
