[Bugs] [Bug 1369933] New: need a way to detect and resolve dangling gfid files

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1369933

            Bug ID: 1369933
           Summary: need a way to detect and resolve dangling gfid files
           Product: GlusterFS
           Version: 3.7.14
         Component: replicate
          Severity: low
          Assignee: bugs at gluster.org
          Reporter: jaco at uls.co.za
                CC: bugs at gluster.org



Description of problem:

I saw a case where a gfid file would end up with a link-count of 1. 
Specifically gluster volume heal ${volname} info reported a number of files
under heal (which didn't heal by themselves).  Upon inspection I found that the
listed gfid's referenced files that had a link count of 1.  I hope that's a
clear enough explanation, but just to be sure:



    bagheera ~ # gluster volume heal mail info            
    Brick uriel.interexcel.co.za:/mnt/gluster/mail
    <gfid:547eaa89-0f80-4507-8d11-cb370bc36eae>
    <gfid:bf518ba9-c2e4-4d9e-bac2-9cf3c32770d1>
    <gfid:266a624a-5788-4cd2-998c-7f727d6b5144>
    <gfid:4a620549-fee1-432b-a865-a6e7bcee1e28>
    <gfid:833618dd-0e21-46fb-812f-6c211c60dcfe>
    <gfid:6f9c0dbd-18f3-4bcd-a018-12afc8676d9f>
    Status: Connected
    Number of entries: 6

    Brick bagheera.interexcel.co.za:/mnt/gluster/mail
    <gfid:2daab67e-56be-455d-a18e-f75e01a5fc75>
    <gfid:94c2261d-df33-4260-9c72-7e61b7c63e36>
    <gfid:2aec1ef1-d95a-41c4-a17d-591be32731f1>
    <gfid:0289814e-91b3-42b8-b3b9-99ab60e81e99>
    <gfid:a9a7e634-760f-4cab-be84-d063dedebf46>
    <gfid:cf585252-e536-4d51-9a27-ef399da1692c>
    <gfid:90ba5183-e187-4a35-874d-aa2c92d47160>
    Status: Connected
    Number of entries: 7

    mail # gluster volume heal mail info | sed -e '1,/^Number/ d' | sed -nre
's/^<gfid:([-0-9a-z]+)>/\1/p' | while read gfid; do stat -c%h\ %n
.glusterfs/${gfid:0:2}/${gfid:2:2}/${gfid}; done
    1 .glusterfs/2d/aa/2daab67e-56be-455d-a18e-f75e01a5fc75
    1 .glusterfs/94/c2/94c2261d-df33-4260-9c72-7e61b7c63e36
    1 .glusterfs/2a/ec/2aec1ef1-d95a-41c4-a17d-591be32731f1
    1 .glusterfs/02/89/0289814e-91b3-42b8-b3b9-99ab60e81e99
    1 .glusterfs/a9/a7/a9a7e634-760f-4cab-be84-d063dedebf46
    1 .glusterfs/cf/58/cf585252-e536-4d51-9a27-ef399da1692c
    1 .glusterfs/90/ba/90ba5183-e187-4a35-874d-aa2c92d47160

    %h - number of hard links

At this point there is basically not much that can be done, one can look at the
content of the file and copy it where it belongs (via the client most likely),
and then ultimately one has to rm that dangling file (which also resolves the
heal entry).

As per suggestion on IRC this situation probably deserves some special
reporting in volume heal ... info (similar to split brain) along with a way of
fixing it, two possible resolutions:

1.  volume heal ${volname} dangling link <gfid> <path>; or
2.  volume heal ${volname} danglink drop <gfid>

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.