[
https://issues.apache.org/jira/browse/HADOOP-14130?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15888876#comment-15888876
]
Mingliang Liu commented on HADOOP-14130:
----------------------------------------
{quote}
I think that's likely to happen with DynamoDB too.
{quote}
I think Steve's last comment should make perfect sense: we add it back if ever
needed in future. Is there any use case that region itself is not enough? We
don't have to expose every feature DDB/AWS provides to our user. Removing
endpoint by now is not "-1 on adding it back later when needed", if I have to
make clear.
{quote}
We discussed this in HADOOP-14090
{quote}
Thanks, I noticed that. I read the discussion and commented. I also mentioned
in this JIRA's description, though missing the link (just updated). The major
conclusion was not to have both.
{quote}
If we decide to have both...
{quote}
I'd -1 if we have both: S3 region, DDB region, DDB endpoint, it's confusing
myself. First, it merits to have two of them if they work just fine: S3 region
and DDB region, or S3 region or DDB endpoint. In the previous implementation,
we chose DDB endpoint because DDBLocal in unit test was a special case as we
used the same DynamoDBClientFactory implementation (for simplicity of initial
effort). Second, the new {{AmazonDynamoDBClientBuilder}} builder we switch to
accepts either region or endpoint + region. So
{quote}
If both are configured, region will be ignored and endpoint will be used
instead"
{quote}
the region can not be ignored. We have to deal with inconsistent endpoint +
region configs.
> Simplify DynamoDBClientFactory for creating Amazon DynamoDB clients
> -------------------------------------------------------------------
>
> Key: HADOOP-14130
> URL: https://issues.apache.org/jira/browse/HADOOP-14130
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: fs/s3
> Reporter: Mingliang Liu
> Assignee: Mingliang Liu
> Attachments: HADOOP-14130-HADOOP-13345.000.patch,
> HADOOP-14130-HADOOP-13345.001.patch
>
>
> So, we are using deprecated {{AmazonDynamoDBClient}} class to create a
> DynamoDB client instead of the recommended builder. We had discussion in
> [HADOOP-14090] for preferring region to endpoints for user to specify the
> DynamoDB region (if associated S3 region is unknown or different). We have
> reported inconsistent behavior if endpoint and S3 region are different in
> [HADOOP-14027]. We also noticed that {{DynamoDBMetadataStore}} may sometimes
> logs nonsense region. And in [HADOOP-13252], we also have feelings that file
> system URI is not needed to create a {{AWSCredentialProvider}}. Resultantly
> we don't need to pass down file system URI for creating a DynamoDB client.
> So this JIRA is to change this, best effort.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
