Spark >= 2.2
Since you 2.2 you can provide format string directly:
import org.apache.spark.sql.functions.to_timestamp
val ts = to_timestamp($"dts", "MM/dd/yyyy HH:mm:ss")
df.withColumn("ts", ts).show(2, false)
// +---+-------------------+-------------------+
// |id |dts |ts |
// +---+-------------------+-------------------+
// |1 |05/26/2016 01:01:01|2016-05-26 01:01:01|
// |2 |#$@#@# |null |
// +---+-------------------+-------------------+
Spark >= 1.6, < 2.2
You can use date processing functions which have been introduced in Spark 1.5. Assuming you have following data:
val df = Seq((1L, "05/26/2016 01:01:01"), (2L, "#$@#@#")).toDF("id", "dts")
You can use unix_timestamp
to parse strings and cast it to timestamp
import org.apache.spark.sql.functions.unix_timestamp
val ts = unix_timestamp($"dts", "MM/dd/yyyy HH:mm:ss").cast("timestamp")
df.withColumn("ts", ts).show(2, false)
// +---+-------------------+---------------------+
// |id |dts |ts |
// +---+-------------------+---------------------+
// |1 |05/26/2016 01:01:01|2016-05-26 01:01:01.0|
// |2 |#$@#@# |null |
// +---+-------------------+---------------------+
As you can see it covers both parsing and error handling. The format string should be compatible with Java SimpleDateFormat
.
Spark >= 1.5, < 1.6
You'll have to use use something like this:
unix_timestamp($"dts", "MM/dd/yyyy HH:mm:ss").cast("double").cast("timestamp")
or
(unix_timestamp($"dts", "MM/dd/yyyy HH:mm:ss") * 1000).cast("timestamp")
due to SPARK-11724.
Spark < 1.5
you should be able to use these with expr
and HiveContext
.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…