Due to the current balancer policy, it is possible for chunks to oscillate(also referred to as ping pong) between the same shards across multiple balancer rounds. For more details about the specific implementation of the balancer/autosplit policy causing this can be found here. Essentially, the main problem is that the balancer assumes that the documents are generally uniform in size because it is too expensive to precisely determine the dimensions of a chunk. Some collections, as seen in HELP-71385 and HELP-73632, may not follow this assumption.

Currently, the two known mitigation options for the ping pong loop are the following:

1. Manually split the chunk.
2. Increase the max chunk size.

The second option was preferred in help cases and successfully resolved the matter. However, this ticket aims to investigate if the balancer can be more proactive about identifying this issue and if the balancer chunk selection policy can be adjusted to prevent it.