[jira] [Created] (CASSANDRA-7887) Skip completing a streamed file if it is identical to one of local files

["Takenori Sato (JIRA)" <ji...@apache.org>]

Takenori Sato created CASSANDRA-7887:
----------------------------------------

             Summary: Skip completing a streamed file if it is identical to one of local files
                 Key: CASSANDRA-7887
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7887
             Project: Cassandra
          Issue Type: Improvement
            Reporter: Takenori Sato


A repair could transfer files that have been done before. Under such a circumstance, those files would get piled up, and a user could experience critical problems such as:

1. overwhelming compaction load
2. disk full

This leads to insufficient repairs, and data inconsistencies.

What about skipping such a file before committing its write(calling SSTableWrite.closeAndOpenReader)? For example, by comparing MD5 hash of a streamed file with those of local files.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)