Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
Currently, if the client reads from all Datanodes in the pipleine fail, the read fails altogether. There may be a case when the container is moved to a new pipeline by the time client reads. In this case, the client should request for a refresh pipeline from OM, and read it again if the new pipeline returned from OM is different.
This behavior is consistent with that of HDFS.
cc msingh / shashikant / hanishakoneru