I think you are right. We are close.

For Python, let's try exploding twice. If we have, [[emailId, date, source], [emailId, date, source], [emailId, date, source]] then let us explode that column out as well so each email ID has its own row.

df = sqlContext.table("owner_final_delta")
import pyspark.sql.functions as F
df.select(F.explode(F.explode(df.contacts.emails).alias("email")).alias("email_ids")).show()

And then for SQL, does this command give you errors as well? I

SELECT from_json(contacts:emails[*], 'array<array<string>>') emails FROM owner_final_delta

I believe emails is an array of an array of strings. I want to see if we can get here first without any errors before digging deeper into the nest.