Logical initial sync can cause a temporary timeseries bucket inconsistency: Buckets contain a version indicating compressed, but the data also includes some uncompressed columns. 

This can happen if the cloning phase of initial sync clones a compressed bucket, but the initial bucket insert is discovered during oplog application. Initial sync has an upsert behavior, which can create a bucket with mixed compressed and uncompressed columns.

Key generation for 2dsphere_bucket index consults the bucket version to determine if it should use the compressed or uncompressed code path and only handles the expected column type.

The problem does not exist on v8.0 as buckets are always in the compressed format.