[ 
https://issues.apache.org/jira/browse/HDFS-10996?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15965109#comment-15965109
 ] 

Andrew Wang commented on HDFS-10996:
------------------------------------

Hi Sammi, thanks for working on this, sorry for the slow review:

bq. I go through the HdfsAdmin and find there is not a single create function 
exposed there. So I'm not sure If I should add a new create API and will that 
benefit its users? Uploaded v5 patch.

Sure, we can tackle this in a separate issue. We're doing some internal testing 
and already found a few issues already regarding the lack of hflush support for 
EC files (e.g. HDFS-11643, and apparently YARN job history too). These could 
potentially be solved by having these apps specify an explicit "replicated" 
policy, which requires a public API.

Some other code review nits, +1 pending these and Jenkins:

* getErasureCodingPolicyByName, we could assert hasReadLock() instead, since 
this method doesn't do any writes
* Need a rebase since ECPolicies have moved to the new 
SystemErasureCodingPolicies class
* testFileLevelECPolicy: nit: "policy should be found" -> "policy should not be 
found"

> Ability to specify per-file EC policy at create time
> ----------------------------------------------------
>
>                 Key: HDFS-10996
>                 URL: https://issues.apache.org/jira/browse/HDFS-10996
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: erasure-coding
>    Affects Versions: 3.0.0-alpha1
>            Reporter: Andrew Wang
>            Assignee: SammiChen
>              Labels: hdfs-ec-3.0-nice-to-have
>         Attachments: HDFS-10996-v1.patch, HDFS-10996-v2.patch, 
> HDFS-10996-v3.patch, HDFS-10996-v4.patch, HDFS-10996-v5.patch
>
>
> Based on discussion in HDFS-10971, it would be useful to specify the EC 
> policy when the file is created. This is useful for situations where app 
> requirements do not map nicely to the current directory-level policies.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org

Reply via email to