使用查找状态更改的位置 lead ，只保留 status 以及 month 与状态更改相对应，并用null屏蔽，否则使用 when(F.col('begin'), F.col('status')) ，并使用 F.last(..., ignorenulls=True) .

import pyspark.sql.functions as F
from pyspark.sql.window import Window

w = Window.partitionBy('user_id').orderBy('month')
begin = F.lead('status').over(w) != F.col('status')
df = df.select('*', begin.alias('begin'))

w = w.rowsBetween(Window.unboundedPreceding, -1)
previous_status_value = F.last(F.when(F.col('begin'), F.col('status')), ignorenulls=True).over(w).alias('previous_status_value')
previous_status_month = F.last(F.when(F.col('begin'), F.col('month')), ignorenulls=True).over(w).alias('previous_status_month ')

df = df.select('*', previous_status_value, previous_status_month).drop('begin').orderBy('user_id', F.col('month').desc())

df.show()
+-------+------+-------+---------------------+----------------------+
|user_id|status|  month|previous_status_value|previous_status_month |
+-------+------+-------+---------------------+----------------------+
|      1|     A|12/2020|                    B|               10/2020|
|      1|     A|11/2020|                    B|               10/2020|
|      1|     B|10/2020|                    A|               08/2020|
|      1|     B|09/2020|                    A|               08/2020|
|      1|     A|08/2020|                    C|               07/2020|
|      1|     C|07/2020|                 null|                  null|
|      2|     A|12/2020|                    B|               09/2020|
|      2|     A|11/2020|                    B|               09/2020|
|      2|     A|10/2020|                    B|               09/2020|
|      2|     B|09/2020|                 null|                  null|
+-------+------+-------+---------------------+----------------------+