Partition of Timestamp column in Dataframes Pyspark - apache-spark

Question

I have a DataFrame in PSspark in the below format
Date Id Name Hours Dno Dname
12/11/2013 1 sam 8 102 It
12/10/2013 2 Ram 7 102 It
11/10/2013 3 Jack 8 103 Accounts
12/11/2013 4 Jim 9 101 Marketing
I want to do partition based on dno and save as table in Hive using Parquet format.
df.write.saveAsTable(
'default.testing', mode='overwrite', partitionBy='Dno', format='parquet')
The query worked fine and created table in Hive with Parquet input.
Now I want to do partitioned based on the year and month of the date column. The timestamp is Unix timestamp
how can we achieve that in PySpark. I have done it in hive but unable to do it PySpark

Answer 1

Spark >= 3.1
Instead of cast use timestamp_seconds
from pyspark.sql.functions import timestamp_seconds
year(timestamp_seconds(col("timestamp")))
Spark < 3.1
Just extract fields you want to use and provide a list of columns as an argument to the partitionBy of the writer. If timestamp is UNIX timestamps expressed in seconds:
df = sc.parallelize([
(1484810378, 1, "sam", 8, 102, "It"),
(1484815300, 2, "ram", 7, 103, "Accounts")
]).toDF(["timestamp", "id", "name", "hours", "dno", "dname"])
add columns:
from pyspark.sql.functions import year, month, col
df_with_year_and_month = (df
.withColumn("year", year(col("timestamp").cast("timestamp")))
.withColumn("month", month(col("timestamp").cast("timestamp"))))
and write:
(df_with_year_and_month
.write
.partitionBy("year", "month")
.mode("overwrite")
.format("parquet")
.saveAsTable("default.testing"))