Github user ouyangxiaochen commented on the issue:
https://github.com/apache/spark/pull/22813
Yes, it happened in our real environment.
The scenario as follows:
Some disk corruption in the production cluster which is normal.
SPARK_LOCAL_DIRS = /data1/bigdata/spark/tmp, disk `data1 `is broken, the
maintenance engineer modified `data1 `to `data2 ` or another. Unfortunately.
the config SPARK_WORK_DIR = /data2/bigdata/spark/tmp, and then we start a
`Thriftserver` process, some temporay folder will be created at the new config
path which is the same as SPARK_WORK_DIR, but when the cleanup cycle time is
reached, the folder created by `Thriftserver` will be removed by
`WorkDirCleanUP`, so it will cause the Beeline and JDBC query to fail.
There is a very extreme situation that the user configures the operating
system directory, which will cause a lot of trouble. So i think add this
condition could reduce some unnecessary risks.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
