Hi Folks,

I wanted to check why spark doesn't create staging dir while doing an
insertInto on partitioned tables. I'm running below example code –
```
spark.sql("set hive.exec.dynamic.partition.mode=nonstrict")

val rdd = sc.parallelize(Seq((1, 5, 1), (2, 1, 2), (4, 4, 3)))
val df = spark.createDataFrame(rdd)
df.write.insertInto("testing_table") // testing table is partitioned on "_1"
```
In this scenario FileOutputCommitter considers table path as output path
and creates temporary folders like
`<some_location>/testing_table/_temporary/0` and then moves them to the
partition location when the job commit happens.

But in-case if multiple parallel apps are inserting into the same
partition, this can cause race condition issues while deleting the
`_temporary` dir. Ideally for each app there should be a unique staging dir
where the job should write its output.

Is there any specific reason for this? or am i missing something here?
Thanks for your time and assistance regarding this!

Kind regards
Sanskar

Reply via email to