HIVE-24838. Reduce FS creation in Warehouse::getDnsPath for object stores

[zeroflag]

Based on its javadoc comment getDnsPath is supposed to change the authority part (ip) of a path to a domain name. To do so it creates a new FileSystem object and gets the default path from it, and uses the authority from the the default path to replace the authority of the input.

There are multiple problems with this. Instantiating a new FileSystem is expensive. Replacing the authority on a blobstore path (s3a) is not needed since s3 uses bucket name instead of a hostname in the path.

Even in HDFS case the original function doesn't do what the original intention might was. The new file system is initialized with the input path so the default FS path is always same as the input path (unless a relative path is used as an input). The only useful thing this function does is transforming a relative path to an absolute path.

For example:

input = /warehouse/tablespace/managed/hive
output = hdfs://namenode-address.site:8020/warehouse/tablespace/managed/hive

But this can be achieved by doing a simple config lookup (FileSystem.getDefaultUri(conf)) there is no need to create a new FS object every time.

[rbalamohan]

Thanks for the patch @zeroflag .

LGTM. +1.

[zhangbutao]

hi @zeroflag , thanks your fix. A late comment: Do you think this optimization applies to hdfs as well?
I think the getDnsPath is just used to convert the relative path to absolute path. FileSystem.getDefaultUri(conf) is enough to get the authority&scheme if the path does not contains authority&scheme.