Re: heal-count vs heal-info and "continual heals"?

Pranith Kumar Karampuri <pkarampu@xxxxxxxxxx> · Tue, 19 Apr 2016 13:11:12 +0530



    On 04/19/2016 12:41 PM, Krutika
      Dhananjay wrote:

    
          On Tue, Apr 19, 2016 at 5:30 AM,
            Lindsay Mathieson <lindsay.mathieson@xxxxxxxxx>
            wrote:

            
               Is it possible to
                confirm the following understandings on my part?

                
                - heal info shows the list of files with uncommitted
                writes across the bricks that would need heals *if*
                a brick went down before the write was committed. 

              
            Yes and no. So heal-info reports both files that truly
              need heal and those that are currently participating in an
              uncommitted transaction. The latter entries are reported
              as "possible undergoing heal". I believe we are going to
              drop the latter category of entries going forward. IIUC
              Pranith is working on something that does that. But I will
              let him confirm that.

            
    Lindsay,

             Heal info must show all the files/dirs that require heal.
    At the moment there is a bug where it says 'possibly undergoing
    heal' even when the file doesn't need heal, for which we sent a fix
    http://review.gluster.org/13873, this is undergoing review and
    should be available in next releases. Once this is in, Your
    understanding is correct. Only thing to add is, the write may fail
    with either brick going down or any other error too like ENOSPACE
    etc.

    
    Pranith

    
                 * additionally it lists files being healed if a
                heal is in process 

                  * writes can be data and metadata. 

                 
                - statics heal-count shows the count of files *actually

                  needing healing*
            
            
            This is correct for the most part. The only remote case
              where it will not report pending heals when it is supposed
              to is if some entries were marked "dirty" (indicated by
              the presence of their index under
              .glusterfs/indices/dirty). So it only considers count of
              entries needing heal in .glusterfs/indices/xattrop but not
              .glusterfs/indices/dirty.

              
            For what it's worth, I realised that we don't really
              need to use locks to examine whether files need heal or
              not if the entries are under indices/xattrop. So if there
              is an entry under .glusterfs/indices/xattrop, it means it
              definitely needs heal. Only the entries under
              .glusterfs/indices/dirty need careful inspection. This
              will significantly enhance heal-info response time. So I
              will

            
            work on getting this change in, maybe for the next .x
              release.

            
    I think with Krutika's fix heal info will be super fast compared to
    what we have now. Please use heal info. If there are any bugs we
    will fix them, it handles all corner cases.

    
    Pranith

    
            -Krutika

            
                So its perfectly ok to see files being listed by heal
                info on a busy cluster, so long as they aren't marked as
                a heal in progress. 

                
                thanks ,

                    -- 
Lindsay Mathieson
                  
              
              _______________________________________________

              Gluster-users mailing list

              Gluster-users@xxxxxxxxxxx

              http://www.gluster.org/mailman/listinfo/gluster-users

            
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://www.gluster.org/mailman/listinfo/gluster-users