[ https://issues.apache.org/jira/browse/HADOOP-11463?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14278448#comment-14278448 ]
Thomas Demoor commented on HADOOP-11463: ---------------------------------------- [~ste...@apache.org] shutDown terminates immediately. In-process transfers are abandoned. For regular uploads, S3 simply discards the object. However, for multi-part uploads, parts that were completely uploaded are stored (and paid for) forever and the multi-part upload is "in progress". The purging functionality alleviates this. If fs.s3a.multipart.purge == true, the constructor of S3AFileSystem aborts all in process multi-part uploads older than fs.s3a.multipart.purge.age seconds (this protects "active" multi-part uploads). [~tedyu] Yeah, that was to be expected. See comment 2 in the description above. > Replace method-local TransferManager object with S3AFileSystem#transfers > ------------------------------------------------------------------------ > > Key: HADOOP-11463 > URL: https://issues.apache.org/jira/browse/HADOOP-11463 > Project: Hadoop Common > Issue Type: Task > Reporter: Ted Yu > Assignee: Ted Yu > Attachments: hadoop-11463-001.patch > > > This is continuation of HADOOP-11446. > The following changes are made according to Thomas Demoor's comments: > 1. Replace method-local TransferManager object with S3AFileSystem#transfers > 2. Do not shutdown TransferManager after purging existing multipart file - > otherwise the current transfer is unable to proceed > 3. Shutdown TransferManager instance in the close method of S3AFileSystem -- This message was sent by Atlassian JIRA (v6.3.4#6332)