[
https://issues.apache.org/jira/browse/HDDS-807?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16784043#comment-16784043
]
Siddharth Wagle edited comment on HDDS-807 at 3/5/19 4:22 AM:
--------------------------------------------------------------
[~bharatviswa] The mock tests do test out the changes that are made in the
patch. IMHO, the test for
{noformat}
org.apache.hadoop.ozone.client.OzoneClientFactory#getRpcClient(java.lang.String,
java.lang.Integer, org.apache.hadoop.conf.Configuration)
{noformat} should not be in the scope of this Jira. I do see that there are no
direct calls that instantiate the RpcClient with OM host and port. We should
open another Jira to test out the RpcClient primed with user-provided host port
as a separate Jira to improve test coverage. If you agree I'd be happy to file
separate Jira for it.
The FITs actually are quite heavy-weight and I deliberately wanted to add the
mock tests to make sure testing a small change like this does not need a
cluster spin up.
was (Author: swagle):
[~bharatviswa] The mock tests do test out the changes that are made in the
patch. IMHO, the test for
{noformat}
org.apache.hadoop.ozone.client.OzoneClientFactory#getRpcClient(java.lang.String,
java.lang.Integer, org.apache.hadoop.conf.Configuration)
{noformat} should not be in the scope of this Jira. I do see that there are no
direct calls that instantiate the RpcClient with OM host and port. We should
open another Jira to test out the RpcClient primed with user-provided host port
as a separate Jira to improve test coverage.
> Period should be an invalid character in bucket names
> -----------------------------------------------------
>
> Key: HDDS-807
> URL: https://issues.apache.org/jira/browse/HDDS-807
> Project: Hadoop Distributed Data Store
> Issue Type: Improvement
> Reporter: Arpit Agarwal
> Assignee: Siddharth Wagle
> Priority: Critical
> Labels: newbie
> Attachments: HDDS-807.01.patch, HDDS-807.02.patch, HDDS-807.03.patch,
> HDDS-807.04.patch
>
>
> ozonefs paths use the following syntax:
> - o3fs://bucket.volume/..
> The OM host and port are read from configuration. We cannot specify a target
> filesystem with a fully qualified path. E.g.
> _o3fs://bucket.volume.om-host.example.com:9862/. Hence we cannot hand a fully
> qualified URL with OM hostname to a client without setting up config files
> beforehand. This is inconvenient. It also means there is no way to perform a
> distcp from one Ozone cluster to another.
> We need a way to support fully qualified paths with OM hostname and port
> _bucket.volume.om-host.example.com_. If we allow periods in bucketnames, then
> such fully qualified paths cannot be parsed unambiguously. However if er
> disallow periods, then we can support all of the following paths
> unambiguously.
> # *o3fs://bucket.volume/key* - The authority has only two period-separated
> components. These must be bucket and volume name respectively.
> # *o3fs://bucket.volume.om-host.example.com/key* - The authority has more
> than two components. The first two must be bucket and volume, the rest must
> be the hostname.
> # *o3fs://bucket.volume.om-host.example.com:5678/key* - Similar to #2,
> except with a port number.
>
> Open question is around HA support. I believe for HA we will have to
> introduce the notion of a _nameservice_, similar to HDFS nameservice. This
> will allow a fourth kind of Ozone URL:
> - *o3fs://bucket.volume.ns1/key* - How do we distinguish this from #3 above?
> One way could be to find if _ns1_ is known as an Ozone nameservice via
> configuration. If so then treat it as the name of an HA service. Else treat
> it as a hostname.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
