Details

    • Type: Improvement Improvement
    • Status: Open
    • Priority: Major Major
    • Resolution: Unresolved
    • Affects Version/s: 2.3.0
    • Fix Version/s: None
    • Component/s: fs
    • Labels:
      None
    • Target Version/s:

      Description

      This is an umbrella JIRA for all the things we have to do to make symlinks production-ready for Hadoop 2.3.

      Note that some of these subtasks are scheduled for 2.1.2 / 2.2, but the overall effort is for 2.3.

        Issue Links

          Activity

          Hide
          Colin P. McCabe added a comment -

          does anyone know if there's a JIRA for hftp symlink support?

          Show
          Colin P. McCabe added a comment - does anyone know if there's a JIRA for hftp symlink support?
          Hide
          Chris Nauroth added a comment -

          FYI, I added HDFS-5303 to the list, reporting that symlinks within snapshots can cause some unexpected behavior.

          Show
          Chris Nauroth added a comment - FYI, I added HDFS-5303 to the list, reporting that symlinks within snapshots can cause some unexpected behavior.
          Hide
          David Ongaro added a comment -

          I'm tired waiting for this feature. I guess by now we can safely say it's complexity was underestimated all along. But in fact I think it was also way too ambitious. Most use cases should be already covered with a "hardlink" like feature. With that I mean just an alias within the namenode to the same files. That should be implementable (I hope) in a transparent way without having to change every HDFS tool/library.

          One of the main features of symlinks is that one can link to other filesystem, but for me it seems this would not even be an option for HDFS, because we expect certain performance characteristics of HDFS. So what would we even get that hardlinks can't do? Hardlinks would also avoid having to think about all these complex problems like link cycles.

          So please retire this and lets get hardlinks rolling. We still could get back to this if the need arises afterwards.

          Show
          David Ongaro added a comment - I'm tired waiting for this feature. I guess by now we can safely say it's complexity was underestimated all along. But in fact I think it was also way too ambitious. Most use cases should be already covered with a "hardlink" like feature. With that I mean just an alias within the namenode to the same files. That should be implementable (I hope) in a transparent way without having to change every HDFS tool/library. One of the main features of symlinks is that one can link to other filesystem, but for me it seems this would not even be an option for HDFS, because we expect certain performance characteristics of HDFS. So what would we even get that hardlinks can't do? Hardlinks would also avoid having to think about all these complex problems like link cycles. So please retire this and lets get hardlinks rolling. We still could get back to this if the need arises afterwards.

            People

            • Assignee:
              Ajith S
              Reporter:
              Colin P. McCabe
            • Votes:
              15 Vote for this issue
              Watchers:
              47 Start watching this issue

              Dates

              • Created:
                Updated:

                Development