[Gluster-users] recommended upgrade procedure from gluster-3.2.7 to gluster-3.5.0

[Todd Pfaff]

thanks, pranith, that was very helpful!

i followed your advice, it ran and completed, and now i'm left with these
results on the removed brick (before commit):

   find /2/scratch/ | wc -l
   83083

   find /2/scratch/ -type f | wc -l
   16

   find /2/scratch/ -type d | wc -l
   70243

   find /2/scratch/ -type l | wc -l
   12824

   find /2/scratch/ ! -type d -a ! -type f | wc -l
   12824

   find /2/scratch/.glusterfs -type l | wc -l
   12824

   find /2/scratch/* | wc -l
   12873

   find /2/scratch/* -type d | wc -l
   12857

so it looks like i have 16 files and 12857 directories left in /2/scratch,
and 12824 links under /2/scratch/.glusterfs/.

my first instinct is to ignore (and remove) the many remaining directories
that are empty and only look closer at those that contain the 16 remaining
files.

can i ignore the links under /2/scratch/.glusterfs?

as for the 16 files that remain, i can migrate them manually if necessary
but i'll first look at all the brick filesystems to see if they already
exist elsewhere in some form.

do you recommend i do anything else?

thanks,
tp


On Fri, 30 May 2014, Pranith Kumar Karampuri wrote:

>
>
> ----- Original Message -----
>> From: "Todd Pfaff" <pfaff at rhpcs.mcmaster.ca>
>> To: gluster-users at gluster.org
>> Sent: Saturday, May 31, 2014 1:58:33 AM
>> Subject: Re: [Gluster-users] recommended upgrade procedure from gluster-3.2.7 to gluster-3.5.0
>>
>> On Sat, 24 May 2014, Todd Pfaff wrote:
>>
>>> I have a gluster distributed volume that has been running nicely with
>>> gluster-3.2.7 for the past two years and I now want to upgrade this to
>>> gluster-3.5.0.
>>>
>>> What is the recommended procedure for such an upgrade?  Is it necessary to
>>> upgrade from 3.2.7 to 3.3 to 3.4 to 3.5, or can I safely transition from
>>> 3.2.7 directly to 3.5.0?
>>
>> nobody responded so i decided to wing it and hope for the best.
>>
>> i also decided to go directly from 3.2.7 to 3.4.3 and not bother with
>> 3.5 yet.
>>
>> the volume is distributed across 13 bricks.  formerly these were in 13
>> nodes, 1 brick per node, but i recently lost one of these nodes.
>> i've moved the brick from the dead node to be a second brick in one of
>> the remaining 12 nodes.  i currently have this state:
>>
>>    gluster volume status
>>    Status of volume: scratch
>>    Gluster process                                 Port    Online  Pid
>>    ------------------------------------------------------------------------------
>>    Brick 172.16.1.1:/1/scratch                     49152   Y       6452
>>    Brick 172.16.1.2:/1/scratch                     49152   Y       10783
>>    Brick 172.16.1.3:/1/scratch                     49152   Y       10164
>>    Brick 172.16.1.4:/1/scratch                     49152   Y       10465
>>    Brick 172.16.1.5:/1/scratch                     49152   Y       10186
>>    Brick 172.16.1.6:/1/scratch                     49152   Y       10388
>>    Brick 172.16.1.7:/1/scratch                     49152   Y       10386
>>    Brick 172.16.1.8:/1/scratch                     49152   Y       10215
>>    Brick 172.16.1.9:/1/scratch                     49152   Y       11059
>>    Brick 172.16.1.10:/1/scratch                    49152   Y       9238
>>    Brick 172.16.1.11:/1/scratch                    49152   Y       9466
>>    Brick 172.16.1.12:/1/scratch                    49152   Y       10777
>>    Brick 172.16.1.1:/2/scratch                     49153   Y       6461
>>
>>
>> what i want to do next is remove Brick 172.16.1.1:/2/scratch and have
>> all files it contains redistributed across the other 12 bricks.
>>
>> what's the correct procedure for this?  is it as simple as:
>>
>>    gluster volume remove-brick scratch 172.16.1.1:/2/scratch start
>>
>> and then wait for all files to be moved off that brick?  or do i also
>> have to do:
>>
>>    gluster volume remove-brick scratch 172.16.1.1:/2/scratch commit
>>
>> and then wait for all files to be moved off that brick?  or do i also
>> have to do something else, such as a rebalance, to cause the files to
>> be moved?
>
> 'gluster volume remove-brick scratch  172.16.1.1:/2/scratch start' does start the process of migrating all the files to the other bricks. You need to observe the progress of the process using 'gluster volume remove-brick scratch  172.16.1.1:/2/scratch status' Once this command says 'completed' You should execute 'gluster volume remove-brick scratch  172.16.1.1:/2/scratch commit' to completely remove this brick from the volume. I am a bit paranoid so I would check that no files are left behind by doing a find on the brick 172.16.1.1:/2/scratch just before issuing the 'commit' :-).
>
> Pranith.
>
>>
>> how do i know when everything has been moved safely to other bricks and
>> the then-empty brick is no longer involved in the cluster?
>>
>> thanks,
>> tp
>>
>> _______________________________________________
>> Gluster-users mailing list
>> Gluster-users at gluster.org
>> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>>
>
>