The data size aware balancing introduced in v6.0.3+ versions can potentially schedule migrations from shards under the optimal data size threshold.

The chunk-based balancing was avoiding that by scheduling migrations only from shards with a sufficiently high number of chunks .

Example
There are 500GB to spread across 5 shards (100GB per shard) currently distributed in the following way:

• Shard0: 200GB
• Shard1: 200GB
• Shard2: 50GB
• Shard3: 40GB
• Shard4: 10GB
• Shard5: 0GB

Next balancing round (currently):

• Shard0 donates to shard5
• Shard1 donates to shard4
• Shard2 donates to shard3 (this is the bug -> there is no real need to donate from shard2 to shard3 because they both need to reach 100GB)

Next balancing round (after this fix):

• Shard0 donates to shard5
• Shard1 donates to shard4