Behavior

When an operation's maxTimeMS / opCtx deadline fires while mongos is sleeping in the IRRL retry backoff, the sleep is cancelled and recordBackoff() (invoked in the .then() callback after waitUntil) never runs. As a result, shardingStatistics.shards.<shard>.totalBackoffTimeMillis is not incremented by the backoff time already elapsed before cancellation, under-reporting the metric.

Discovery

Discovered while writing the test in SERVER-127631 (PR #57098), where totalBackoffTimeMillis is asserted to be 0 despite time genuinely spent in the backoff sleep.

Low priority.