[SPARK-15763][SQL] Support DELETE FILE command natively

[kevinyu98]

What changes were proposed in this pull request?

Hive supports these cli commands to manage the resource Hive Doc :
ADD/DELETE (FILE(s)<filepath..>|JAR(s) <jarpath..>)
LIST (FILE(S) [filepath ...] | JAR(S) [jarpath ...])

but SPARK only supports two commands
ADD (FILE <filepath> | JAR <jarpath>)
LIST (FILE(S) [filepath ...] | JAR(S) [jarpath ...]) for now.

This PR is to add the DELETE FILE command into Spark SQL and I will submit another PR for the DELETE JAR(s).

DELETE FILE <filepath>

Example:

DELETE FILE

scala> spark.sql("add file /Users/qianyangyu/myfile.txt")
res0: org.apache.spark.sql.DataFrame = []

scala> spark.sql("add file /Users/qianyangyu/myfile2.txt")
res1: org.apache.spark.sql.DataFrame = []

scala> spark.sql("list file")
res2: org.apache.spark.sql.DataFrame = [Results: string]

scala> spark.sql("list file").show(false)
+----------------------------------+
|Results                           |
+----------------------------------+
|file:/Users/qianyangyu/myfile2.txt|
|file:/Users/qianyangyu/myfile.txt |
+----------------------------------+
scala> spark.sql("delete file /Users/qianyangyu/myfile.txt")
res4: org.apache.spark.sql.DataFrame = []

scala> spark.sql("list file").show(false)
+----------------------------------+
|Results                           |
+----------------------------------+
|file:/Users/qianyangyu/myfile2.txt|
+----------------------------------+


scala> spark.sql("delete file /Users/qianyangyu/myfile2.txt")
res6: org.apache.spark.sql.DataFrame = []

scala> spark.sql("list file").show(false)
+-------+
|Results|
+-------+
+-------+

How was this patch tested?

Add test cases in Spark-SQL SPARK-Shell and SparkContext suites.

(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)

[AmplabJenkins]

Can one of the admins verify this patch?

[rxin]

this is fairly confusing -- i'd assume this is actually deleting the path given.

[kevinyu98]

Hi Reynold: Thanks very much for reviewing the code.
yes, it is deleting the path from the addedFile hashmap, the path will be generated as key and stored in the map.
The addFile use this logical to generate the key and stored in the hashmap, so in order to find the same key, I have to use the same logical to generate the key.
For example:
for this local file, the addFile will generate a 'file' in front of the path.

spark.sql("add file /Users/qianyangyu/myfile.txt")

scala> spark.sql("list file").show(false)
+----------------------------------+
|Results |
+----------------------------------+
|file:/Users/qianyangyu/myfile2.txt|
|file:/Users/qianyangyu/myfile.txt |
+----------------------------------+

but for the remote location file, it will just take the path.

scala> spark.sql("add file hdfs://bdavm009.svl.ibm.com:8020/tmp/test.txt")
res17: org.apache.spark.sql.DataFrame = []

scala> spark.sql("list file").show(false)
+---------------------------------------------+
|Results |
+---------------------------------------------+
|file:/Users/qianyangyu/myfile.txt |
|hdfs://bdavm009.svl.ibm.com:8020/tmp/test.txt|
+---------------------------------------------+

if the command is issued from the worker node and add local file, the path will be added into the NettyStreamManager's hashmap and using that environment's path as key to store in the addedFiles.

[kevinyu98]

I have updated the deleteFile comments to make it more clear. Thanks for reviewing.

[vanzin]

@kevinyu98 Could you update the PR and fix merge conflicts? Thanks

[kevinyu98]

@vanzin Hello Marcelo: I am so sorry that I didn't notice your update. I have fix the merge conflicts and can you help review it? Thanks.

[gatorsmile]

@kevinyu98 Can you please close it? It seems like there is not a lot of interest in adding this functionality natively in Spark. If anybody wants this feature, we can reopen it later?

[kevinyu98]

sure

[gatorsmile]

We are closing it due to inactivity. please do reopen if you want to push it forward. Thanks!