You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-issues@hadoop.apache.org by "Chris Nauroth (JIRA)" <ji...@apache.org> on 2015/12/22 18:37:46 UTC
[jira] [Commented] (HADOOP-12667) s3a: Support createNonRecursive
API
[ https://issues.apache.org/jira/browse/HADOOP-12667?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15068449#comment-15068449 ]
Chris Nauroth commented on HADOOP-12667:
----------------------------------------
If the goal is to support HBase's WAL management, then another aspect is that HBase expects atomicity from this call. Atomicity would guarantee that for 2 concurrent threads/processes calling {{createNonRecursive}}, one of the callers succeeds and the other fails. HBASE-11045 has some great discussion of the expected semantics, especially this comment:
https://issues.apache.org/jira/browse/HBASE-11045?focusedCommentId=13977198&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13977198
This means that an implementation that just checks for existence before allowing the create to proceed is insufficient. There would be a race condition allowing 2 concurrent callers to think the path exists and proceed with their creates.
In hadoop-azure, this problem is solved by making concurrent callers acquire a mutually exclusive lease on the blob before doing the create. Blob leases are a feature of Azure Storage. I'm not aware of any lease functionality like this available in S3.
> s3a: Support createNonRecursive API
> -----------------------------------
>
> Key: HADOOP-12667
> URL: https://issues.apache.org/jira/browse/HADOOP-12667
> Project: Hadoop Common
> Issue Type: Bug
> Components: fs/s3
> Reporter: Sean Mackrory
> Assignee: Sean Mackrory
> Attachments: HADOOP-12667.001.patch
>
>
> HBase and other clients rely on the createNonRecursive API, which was recently un-deprecated. S3A currently does not support it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)