无法从 Databricks 笔记本的 Azure 存储容器中删除目录

Question

I'm trying to delete empty directories from Azure storage container which mounted to my DBFS

I'm able to list all directories which has no files.

%sh
find /dbfs/mnt/test/logs/2021 -empty -type d

Result:

/dbfs/mnt/test/logs/2021/02/12
/dbfs/mnt/test/logs/2021/02/15
/dbfs/mnt/test/logs/2021/02/16

But when I try to delete them, it is failing with Resource temporary unavailable.

%sh
find /dbfs/mnt/test/logs/ -type d -exec rmdir {} \; 

Result:

rmdir: failed to remove '/dbfs/mnt/test/logs/': Directory not empty
rmdir: failed to remove '/dbfs/mnt/test/logs/2021': Directory not empty
rmdir: failed to remove '/dbfs/mnt/test/logs/2021/02': Directory not empty
rmdir: failed to remove '/dbfs/mnt/test/logs/2021/02/12': Resource temporarily unavailable

I'm able to successfully remove files older than certain days.. removing direcotry is not working. (Below command to remove files working

%sh
find /dbfs/mnt/test/logs/ -name "*.log" -type f -mtime +5 -exec rm -f {} \; 

Answer 1

First thing to remember - DBFS is an abstraction over the cloud blob storage, where there is no real directories - they are just prefixes that are used to organize data. And if you do %sh ls -ls /dbfs/mnt/test/logs/ you may notice that all directories will have the same timestamp, and it could be the recent one - I don't remember out the head how it's calculated. Only files have the timestamp.

So if you need to reliably remove directories, it's better to use dbutils.fs.rm('/mnt/test/logs/', True) (in Python, or similar in the Scala) to remove directory recursively (see docs ). But there are limitations, like there is no support for wildcards, etc., so you need to generate a list of directories to delete, and do the deletion.