还可以将interval与expr一起使用。

import pyspark.sql.functions as F
df = spark.createDataFrame(
    [
        (1, '2020-07-13 17:29:36')
    ], 
    [
        'id', 'time'
    ]
)
df.withColumn(
    'time', 
    F.col('time').cast('timestamp')
).withColumn(
    'timediff', 
    (
        F.col('time') -  F.expr('INTERVAL 1 milliseconds')
    ).cast('timestamp') 
).show(truncate=False)

通过使用double type，您可以做到这一点。

import pyspark.sql.functions as f
df = spark.createDataFrame([(1, '2020-07-13 17:29:36')], ['id', 'time'])
df.withColumn('time', f.to_timestamp('time', 'yyyy-MM-dd HH:mm:ss')) \
  .withColumn('timediff', (f.col('time').cast('double') - f.lit(0.001)).cast('timestamp')) \
  .show(10, False)
+---+-------------------+-----------------------+
|id |time               |timediff               |
+---+-------------------+-----------------------+
|1  |2020-07-13 17:29:36|2020-07-13 17:29:35.999|
+---+-------------------+-----------------------+

你可以用 pyspark.sql.functions.expr 减去 INTERVAL 1 milliseconds ```
from pyspark.sql.functions import expr

df = spark.createDataFrame([('2020-07-13 17:29:36',)], ['time'])
df = df.withColumn('time2', expr("time - INTERVAL 1 milliseconds"))
df.show(truncate=False)

+-------------------+-----------------------+

|time |time2 |

+-------------------+-----------------------+

|2020-07-13 17:29:36|2020-07-13 17:29:35.999|

+-------------------+-----------------------+

即使 `time` 是这种格式的字符串，spark将为您进行隐式转换。

df.printSchema()

root

|-- time: string (nullable = true)

|-- time2: string (nullable = true)