How does Task.WhenAll behave when multiple tasks fail?

In production I avoid hiding fan-out failure details. I usually keep references to the individual tasks, await WhenAll, and on failure inspect each task's status so logs and telemetry show exactly which dependency failed. I also decide whether cancellation of one operation should cancel the others, whether partial results are safe, and whether retries should happen per dependency. The risk is building a parallel flow that looks elegant but gives poor incident visibility when several dependencies degrade at once.