Currently, when the front-end sends only the `limit` parameter, the admin server attempts to consume at most `limit` records from the Kafka broker(s). The starting offset for the consumer is (for each partition) calculated as `endOffset` - `limit`. When message reside is more than one partition, the messages from partitions other than the first may not appear in the result set, even if they are newer than some messages from the earlier/lower partitions.

Without consuming the messages, it is not known which partition holds the latest messages or how the latest messages are distributed across the partitions.

Possible solutions:

1. Calculate the starting offset as

   endOffset - (limit / partitionCount)

   . This may result in fewer than `limit` records returned.

2. Fetch up to `partitionCount` * `limit` records and return up to `limit` records that are chronologically the latest.