s3mper and EMRFS

[amiorin]

Is EMRFS a fork a s3mper?

I tried s3mper with hadoop 2.7.1 :

• it looks it doesn't support s3a, but only s3n
• the dynamodb table was created in Virginia, is it possible to override the aws region?
• is there anybody using it with spark or presto?

Thx!

[danielcweeks]

EMRFS was implemented internal to EMR's filesystem after s3mper was released with many of the same concepts and features. Overall, the EMRFS implementation is more complete in that it covers all operations whereas s3mper only covers create, delete, and list. The underlying implementation appears to be different at least in how the metadata is stored.

• For s3a filesystem, you need to update the aop.xml file to point to the s3a classes (any implementation of the FileSystem interface can be used).
• There is currently no property for region, but could be added easily to DynamoDBMetastore.java when creating the client.
• I've had multiple people ask about integration, but don't know if they are actively using it. It will work out of the box for spark, but presto uses a new list method not currently covered by s3mper.

Overall, if you have access to EMRFS, I would recommend you use that in lieu of s3mper as it should be more complete and will address consistency issues that s3mper was never intended to resolve.

-Dan