[
https://issues.apache.org/jira/browse/HADOOP-18889?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17767265#comment-17767265
]
Steve Loughran commented on HADOOP-18889:
-----------------------------------------
+500 errors are retried in the aws sdk with the usual backoff; if a store impl
is rejecting an s3 call then s3a retry policy currently treat as a comms error
and retries too (see https://github.com/aws/aws-sdk-java-v2/issues/3483 ).
plan: failfast.
+reject precondition (512) and unsupported media type. should probably look at
the other 500x calls too and treat all as unrecoverable and map to our 500
error. simpler than adding new ones. but we should really have server-side
change tracking to downgrade on a 512 after a warn once, as it would avoid
people having to turn it off explicitly on ozone etc. Or at least we report the
failure better, even though it'll surface on the first get in a stream (oh, and
what about prefetching? we should just give up fetching at that point....)
> S3A: V2 SDK client does not work with third-party store
> -------------------------------------------------------
>
> Key: HADOOP-18889
> URL: https://issues.apache.org/jira/browse/HADOOP-18889
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: fs/s3
> Affects Versions: 3.4.0
> Reporter: Steve Loughran
> Assignee: Steve Loughran
> Priority: Critical
>
> testing against an external store without specifying region now blows up
> because the region is queried off eu-west-1.
> What are we do to here? require the region setting *which wasn't needed
> before? what even region do we provide for third party stores?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
