[ 
https://issues.apache.org/jira/browse/BIGTOP-1221?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13915349#comment-13915349
 ] 

jay vyas commented on BIGTOP-1221:
----------------------------------

Yes, thats why this patch is important :)  

FUSE is an interface, and we should test it as such.  

If we dont want to run certain FUSE tests in HDFS, then we can specify that as 
a configuration parameter.  Creating a special unit test for every FileSystem 
implementation (HDFS,Kosmos,GlusterFileSystem,S3FileSystem,....)  is a really 
massive undertaking.

[~mgrover]  ... is this convincing you at all ?  If not I'm okay changing the 
architecture.  But the key goal here is to have a solid HCFS FUSE test that 
confirms that distributed file system implementations of the FUSE interface 
work correctly.

> Expand and updated FUSE tests
> -----------------------------
>
>                 Key: BIGTOP-1221
>                 URL: https://issues.apache.org/jira/browse/BIGTOP-1221
>             Project: Bigtop
>          Issue Type: Bug
>            Reporter: jay vyas
>            Assignee: jay vyas
>         Attachments: BIGTOP-1221.patch
>
>
> The BigTop TestFuseDFS need 
> - tests for consistency after individual writes : consistency in FUSE mounts 
> is important since there are often associated multitenant workloads.  Writes 
> should result in immediate recognition of a file. 
> - Should use HCFS semantics (That way we can confidently  use and maintain 
> this for alternative file system implementations).
> - should use logging instead of System.out.  That way all the logs can be 
> picked up from the same place in case we change the logging framework.  
> Logging in trace mode, in any case, will allow us to see everything that the 
> ITest shell does, and so might as well use it for other logs.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Reply via email to