[Gluster-devel] [Gluster-users] Self heal files

jayakrishnan mm jayakrishnan.mm at gmail.com
Tue Feb 14 11:28:55 UTC 2017


On Feb 14, 2017 6:58 PM, "Krutika Dhananjay" <kdhananj at redhat.com> wrote:



On Tue, Feb 14, 2017 at 1:01 PM, jayakrishnan mm <jayakrishnan.mm at gmail.com>
wrote:

>
>
>
>
>
> On Mon, Feb 13, 2017 at 7:07 PM, Krutika Dhananjay <kdhananj at redhat.com>
> wrote:
>
>> Hi JK,
>>
>> On Mon, Feb 13, 2017 at 1:06 PM, jayakrishnan mm <
>> jayakrishnan.mm at gmail.com> wrote:
>>
>>> Hi Krutika,
>>> Could you pls. tell me what is the meaning of base file name ? I mean
>>> the part after xattrop-* what do  this number signify ?
>>>
>>> Ex: xattrop-*a321b856-05b3-48d3-**a393-805be83c6b73 . *Is this gfid  of
>>>  some file ?
>>>
>>
>> No. *a321b856-05b3-48d3-**a393-805be83c6b73 *is a randomly UUID.
>> So in order to not consume one on-disk inode per index, index translator
>> creates this one base file called xattrop-xxxxxxxxx....xxx of size 0 bytes
>> where xxxxxxxx.....xxx is a randomly generated UUID and all indices that
>> need to be created to signify that certain gfids need heal will be
>> hard-linked to
>> this base file. So the number of inodes consumed will remain 1,
>> irrespective of
>> whether 10 files need heal or a 100 or 1000 or even a million.
>> You can do an `ls -li` under xattrop directory where a couple of files
>> need heal
>> to see it for yourself.
>>
>
> Understood. So all  files will have gfid entries inside .glusterfs/xx/yy ,
> where xx/yy  is the initial part of gfid.
> Only those need healing are kept under xattrop directory, and in order to
> save inodes, they are hard linked to a base file .
> And  self heal daemon can traverse  this directory later  for the purpose
> of healing.
>

The hard links under .glusterfs/xx/yy are not to be confused with the
indices under indices/xattrop. Even files that *need* heal will have
hard links to the original file containing user-written data that resides
under its normal parent dir, under .glusterfs/xx/yy.
GFID being a unique property of every inode, the indices under xattrop are
named after the gfid of individual files.

>
> I have encountered some extra gfids
>
> *Volume Name: rep-vol*
> *Type: Replicate*
> *Volume ID: 8667044d-b75e-4fc0-9ae8-2d2b39b8558f*
> *Status: Started*
> *Number of Bricks: 1 x 2 = 2*
> *Transport-type: tcp*
> *Bricks:*
> *Brick1: 192.168.36.220:/home/user/gluster/rep-brick1*
> *Brick2: 192.168.36.220:/home/user/gluster/rep-brick2*
> *Options Reconfigured:*
> *performance.readdir-ahead: on*
>
> *I kill  brick1 process, so that status is as below*
>
> Status of volume: rep-vol
> Gluster process                             TCP Port  RDMA Port  Online
>  Pid
> ------------------------------------------------------------
> ------------------
> Brick 192.168.36.220:/home/user/gluster/rep
> -brick1                                     N/A       N/A        N
> N/A
> Brick 192.168.36.220:/home/user/gluster/rep
> -brick2                                     49221     0          Y
> 24925
> NFS Server on localhost                     N/A       N/A        N
> N/A
> Self-heal Daemon on localhost               N/A       N/A        Y
> 24954
>
> Task Status of Volume rep-vol
> ------------------------------------------------------------
> ------------------
> There are no active volume tasks
>
> *Copy  wish.txt  to /mnt/gluster/rep (mount  point)*
>
> *root at dhcp-192-168-36-220:/home/user/gluster/rep-brick2/.glusterfs/indices/xattrop#
> ls -li*
> *total 0*
> *3670235 ---------- 3 root root 0 Feb  9 07:04
> 00000000-0000-0000-0000-000000000001*
> *3670235 ---------- 3 root root 0 Feb  9 07:04
> a3e23755-4ec6-42d2-ac2c-ad4bd682cdbd*
> *3670235 ---------- 3 root root 0 Feb  9 07:04
> xattrop-8263faed-cba8-4738-9197-93e4e7e103ff*
>
> *As expected.*
>
> *Now I create  another file  test.txt  on the mount point.*
>
> *root at dhcp-192-168-36-220:/home/user/gluster/rep-brick2/.glusterfs/indices/xattrop#
> ls -li*
> *total 0*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> 00000000-0000-0000-0000-000000000001*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> 571ca3f1-5c1b-426d-990e-191aa62ea9c4*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> 6b7f7823-b864-4f48-8a07-7c073f8d2ef5*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> 7b4e97fc-8734-4dba-a72a-a750a22abd2d*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> 96587bb0-2ff8-4d97-8470-69bb48be9fd2*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> a3e23755-4ec6-42d2-ac2c-ad4bd682cdbd*
> *3670235 ---------- 7 root root 0 Feb  9 07:04
> xattrop-8263faed-cba8-4738-9197-93e4e7e103ff*
>
> root at dhcp-192-168-36-220:/home/user/gluster/rep-brick2/.glusterfs/indices/xattrop#
> getfattr -d -e hex -m . ../../../wish.txt  | grep gfid
> trusted.gfid=*0xa3e237554ec642d2ac2cad4bd682cdbd*
> root at dhcp-192-168-36-220:/home/user/gluster/rep-brick2/.glusterfs/indices/xattrop#
> getfattr -d -e hex -m . ../../../test.txt  | grep gfid
> trusted.gfid=*0x571ca3f15c1b426d990e191aa62ea9c4*
>
>
> *Why  some extra  gfids?*
>

Did you try converting those extra gfids to path and figure out what
entries those are?
If you haven't, could you please do that and get back?

-Krutika

I will check and let you know.  Meanwhile,  i would like to know  what
priv->pending_key refers to.  I know priv->pending refers to failed nodes.
Regards
JK




>
> Regards
> JK
>
>
>
>>
>> -Krutika
>>
>>
>>> Regards
>>> JK
>>>
>>>
>>> On Tue, Apr 19, 2016 at 12:31 PM, jayakrishnan mm <
>>> jayakrishnan.mm at gmail.com> wrote:
>>>
>>>> OK.
>>>> On Apr 19, 2016 11:25 AM, "Krutika Dhananjay" <kdhananj at redhat.com>
>>>> wrote:
>>>>
>>>>> The hardlink will be removed, yes, but the base file will stay.
>>>>>
>>>>> -Krutika
>>>>>
>>>>> On Tue, Apr 19, 2016 at 8:31 AM, jayakrishnan mm <
>>>>> jayakrishnan.mm at gmail.com> wrote:
>>>>>
>>>>>> Hi,
>>>>>> Is the hardlink  not  removed, after  healing  is  done ?
>>>>>>
>>>>>> --JK
>>>>>>
>>>>>>
>>>>>> On Mon, Apr 18, 2016 at 6:22 PM, Krutika Dhananjay <
>>>>>> kdhananj at redhat.com> wrote:
>>>>>>
>>>>>>> That's just a base file that all the gfid files are hard-linked to.
>>>>>>> Since it is pointless to consume one inode for each gfid that needs
>>>>>>> a heal, we use a base file
>>>>>>> with an identifiable name (xattrop-*) and then hard-link the actual
>>>>>>> gfid files representing pointers for heal
>>>>>>> to this file.
>>>>>>>
>>>>>>> -Krutika
>>>>>>>
>>>>>>> On Mon, Apr 18, 2016 at 11:00 AM, jayakrishnan mm <
>>>>>>> jayakrishnan.mm at gmail.com> wrote:
>>>>>>>
>>>>>>>> Hi,
>>>>>>>> Self  healing daemon refers  to .glusterfs/indices/xattrop
>>>>>>>> directory  to see the  files  which  are  to  be  healed, and  these  dir
>>>>>>>> should  contain gfids  of those  files.
>>>>>>>>
>>>>>>>> I see  some  other ids  also which are  prefixed  with  xattrop-  ,
>>>>>>>> for example :
>>>>>>>>
>>>>>>>> root at ad3:/data/ssd/dsi-ec8-brick/.glusterfs/indices/xattrop# ll
>>>>>>>> total 8
>>>>>>>> drw------- 2 root root 4096 Apr  8 10:53 ./
>>>>>>>> drw------- 3 root root 4096 Apr  8 10:53 ../
>>>>>>>> ---------- 1 root root    0 Apr  8 10:53
>>>>>>>> xattrop-a321b856-05b3-48d3-a393-805be83c6b73
>>>>>>>>
>>>>>>>> What  is the  meaning ?
>>>>>>>>
>>>>>>>>
>>>>>>>> Best Regards
>>>>>>>> JK
>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> Gluster-users mailing list
>>>>>>>> Gluster-users at gluster.org
>>>>>>>> http://www.gluster.org/mailman/listinfo/gluster-users
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-devel/attachments/20170214/89a88dbe/attachment-0001.html>


More information about the Gluster-devel mailing list