Function diskless_retention_prefix 

pub fn diskless_retention_prefix(
    max_timestamps: &[i64],
    byte_lens: &[u64],
    last_offsets: &[i64],
    policy: DisklessRetentionPolicy,
) -> usize

Select the oldest contiguous prefix of one diskless partition’s committed WAL index ranges that retention allows to expire.

Kafka runs three walks oldest first, each stopping at the first segment it must keep: the log-start-offset walk, then the size walk over what is left, then the time walk over what is left after that. The first two predicates only ever hold on a prefix (the size one because sizes are nonnegative, the floor one because the host supplies ranges in offset order), so the three walks together expire exactly the longest prefix on which one of the three holds at every range. That prefix is what this returns, capped one short of the newest range; see expirable.

The ranges arrive oldest first, one entry per index range, and the three slices are parallel.

The newest range never expires. Kafka keeps the active segment for the same reason, and here it is also what keeps the flusher’s flushed_frontier pointing past the last flushed offset: an empty index would send the next tick back to the local log start and re-upload a prefix the bucket holds.