Destination dependent sharding feature is very much needed in order to maintain manageable files sizes and file counts in google storage especially when data volumes are very large.
Current implementation doesn't allow that (per documentation ) https://beam.apache.org/releases/javadoc/2.4.0/org/apache/beam/sdk/io/FileIO.html
Note that currently sharding can not be destination-dependent: every window/pane for every destination will use the same number of shards specified via FileIO.Write.withNumShards(int) or FileIO.Write.withSharding(org.apache.beam.sdk.transforms.PTransform<org.apache.beam.sdk.values.PCollection<UserT>, org.apache.beam.sdk.values.PCollectionView<java.lang.Integer>>).
**We use it as follows and end up with either very small or very large files per destination in the same window. Large files are not possible to open/process and small files clutter the bucket.