View Source

Background

Kylin will generate temporary files in HDFS during the cube building; Besides, when purge/drop/merge cubes, some parquet files may be left in HDFS and will no longer be queried; Although Kylin has started to do some automated garbage collection, it might not cover all cases; You can do an offline storage cleanup periodically.

Directory tree structure under Kylin 4.0 's working dir

Working Dir(ROOT)

{PROJECT_NAME} [managed by tool]
- parquet
  - {CUBE_NAME} [managed by tool]
    - {SEGMENT_NAME} [managed by tool]
      - {CUBOID_ID}
        parquet files
- spark_log
  - driver
    - {JOB_ID}
      - drivers' log of cubing job
  - executor
    - {JOB_ID}
      - executors' log of cubing job
- dict/global_dict [managed by tool]
  - {CUBE_NAME}
    - {COLUMN_NAME}
      - dict files
- table_snapshot [managed by tool]
  - {SCHEMA_NAME.TABLE_NAME}
    - {JOB_ID}
      - parquet files
- job_tmp [managed by tool]
  - {JOB_ID}
    - TBD
cube_statistics
- {CUBE_NAME}
  - {JOB_ID}
    - seq file of cuboid 's HLL
_sparder_log
- {DATE}
  - executors 's log of query job
resources-jdbc
- TBD

Summary

In above directory tree, the directory which end with "managed by tool" means StorageCleanupJob will try to check and delete useless files under these directory.

For directory table_snapshot, dict/global_dict, parquet/{CUBE_NAME}, parquet/{CUBE_NAME}/{SEGMENT_NAME} , Kylin will mark files which is unreferenced and stale(by checking last modified time) as garbage.

For directory job_tmp, Kylin will only check last modified time.

How to use

Option Table

Option	Data Type	Default Value	Comment
delete	Boolean	false	Boolean, whether or not to do real delete operation. Default value is false, means a dry run.
cleanupTableSnapshot	Boolean	true	Boolean, whether or not to delete unreferenced snapshot files. Default value is true .
cleanupGlobalDict	Boolean	true	Boolean, whether or not to delete unreferenced global dict files. Default value is true .
cleanupJobTmp	Boolean	false	Boolean, whether or not to delete job tmp files. Default value is false .
cleanupThreshold	Integer	168	Integer, used to specific delete unreferenced storage that have not been modified before how many hours (recent files are protected). Default value is 168 hours.

List help information

[root@cdh-master apache-kylin-4.0.0-SNAPSHOT-bin]# bin/kylin.sh org.apache.kylin.tool.StorageCleanupJob -help
Retrieving hive dependency...
Retrieving hadoop conf dir...
Retrieving Spark dependency...
...
Running org.apache.kylin.rest.job.StorageCleanupJob -help
usage: org.apache.kylin.rest.job.StorageCleanupJob
-cleanupGlobalDict <cleanupGlobalDict> Boolean, whether or not to
delete unreferenced global
dict files. Default value
is true .
-cleanupJobTmp <cleanupJobTmp> Boolean, whether or not to
delete job tmp files.
Default value is false .
-cleanupTableSnapshot <cleanupTableSnapshot> Boolean, whether or not to
delete unreferenced
snapshot files. Default
value is true .
-cleanupThreshold <cleanupThreshold> Integer, used to specific
delete unreferenced
storage that have not been
modified before how many
hours (recent files are
protected). Default value
is 168 hours.
-delete <delete> Boolean, whether or not to
do real delete operation.
Default value is false,
means a dry run.

List directory which to be deleted

bin/kylin.sh org.apache.kylin.tool.StorageCleanupJob

Deleted them after confirm

bin/kylin.sh org.apache.kylin.tool.StorageCleanupJob --delete true

Only delete stale job_tmp and unreferenced cuboid files

bin/kylin.sh org.apache.kylin.tool.StorageCleanupJob --delete true \
 --cleanupJobTmp ture -cleanupTableSnapshot false \
 -cleanupGlobalDict false --cleanupThreshold 24