arp7 commented on pull request #1149:
URL: https://github.com/apache/hadoop-ozone/pull/1149#issuecomment-665731402
Sorry I didn't get a chance to follow the discussion in detail. Does the PR
summary need an update to describe the new proposal? If so could you please
update it.
arp7 commented on pull request #1149:
URL: https://github.com/apache/hadoop-ozone/pull/1149#issuecomment-656230263
So iiuc the main convenience argument is that we can type paths like
`/user/bob/foo` instead of `o3fs://bucket.volume.om/user/bob/foo` right?
I think some of the
arp7 commented on pull request #1149:
URL: https://github.com/apache/hadoop-ozone/pull/1149#issuecomment-656221254
Yes one risk is when people code up their apps or scripts to use implicit
paths. Then we add a second cluster and now their paths that used to work
previously don't work
arp7 commented on pull request #1149:
URL: https://github.com/apache/hadoop-ozone/pull/1149#issuecomment-655721161
bq. If the user has both remote and local (or multiple federated cluster) it
means that we have two service ids.
We want to support client-less config of HA eventually
arp7 commented on pull request #1149:
URL: https://github.com/apache/hadoop-ozone/pull/1149#issuecomment-655129013
Yeah this was a conscious choice. I feel usage of unqualified paths in HDFS
can be ambiguous and error-prone when multiple clusters/federation are involved.