I'm currently working with mongodb, which is schemaless. In my case I need to gather most possible schema in the collection and compare it with the current schema which is recorded by datahub. So how do I can directly query the schema of the dataset in datahub?

You can query for the schemaMetadata aspect and use the fields in it for comparing (see https://datahubproject.io/docs/cli/#get). Alternately, you could ingest the most-possible schema for the dataset and use the timeline CLI or API to get the schema difference.