Re: Failed file system

"Andres E. Moya" <amoya@xxxxxxxxxxxxxxxxx> · Wed, 3 Aug 2016 15:39:35 -0400 (EDT)

Does anyone else have input?

we are currently only running off 1 node and one node is offline in replicate brick.

we are not experiencing any downtime because the 1 node is up.

I do not understand which is the best way to bring up a second node.

Do we just re create a file system on the node that is down and the mount points and allow gluster to heal( my concern with this is whether the node that is down will some how take precedence and wipe out the data on the healthy node instead of vice versa)

Or do we fully wipe out the config on the node that is down, re create the file system and re add the node that is down into gluster using the add brick command replica 3, and then wait for it to heal then run the remove brick command for the failed brick

which would be the safest and easiest to accomplish

thanks for any input

From: "Leno Vo" <lenovolastname@xxxxxxxxx>
To: "Andres E. Moya" <amoya@xxxxxxxxxxxxxxxxx>
Cc: "gluster-users" <gluster-users@xxxxxxxxxxx>
Sent: Tuesday, August 2, 2016 6:45:27 PM
Subject: Re:  Failed file system

if you don't want any downtime (in the case that your node 2 really die), you have to create a new gluster san (if you have the resources of course, 3 nodes as much as possible this time), and then just migrate your vms (or files), therefore no downtime but you have to cross your finger that the only node will not die too...  also without sharding the vm migration especially an rdp one, will be slow access from users till it migrated.

you have to start testing sharding, it's fast and cool...

    On Tuesday, August 2, 2016 2:51 PM, Andres E. Moya <amoya@xxxxxxxxxxxxxxxxx> wrote:

 couldnt we just add a new server by

gluster peer probe
gluster volume add-brick replica 3 (will this command succeed with 1 current failed brick?)

let it heal, then 

gluster volume remove remove-brick
From: "Leno Vo" <lenovolastname@xxxxxxxxx>
To: "Andres E. Moya" <amoya@xxxxxxxxxxxxxxxxx>, "gluster-users" <gluster-users@xxxxxxxxxxx>
Sent: Tuesday, August 2, 2016 1:26:42 PM
Subject: Re:  Failed file system

you need to have a downtime to recreate the second node, two nodes is actually not good for production and you should have put raid 1 or raid 5 as your gluster storage, when you recreate the second node you might try running some VMs that need to be up and rest of vm need to be down but stop all backup and if you have replication, stop it too.  if you have 1G nic, 2cpu and less 8Gram, then i suggest all turn off the VMs during recreation of second node. someone said if you have sharding with 3.7.x, maybe some vip vm can be up...

if it just a filesystem, then just turn off the backup service until you recreate the second node. depending on your resources and how big is your storage, it might be hours to recreate it and even days...

here's my process on recreating the second or third node (copied and modifed from the net),

#make sure partition is already added!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
This procedure is for replacing a failed server, IF your newly installed server has the same hostname as the failed one:

(If your new server will have a different hostname, see this article instead.)

For purposes of this example, the server that crashed will be server3 and the other servers will be server1 and server2

On both server1 and server2, make sure hostname server3 resolves to the correct IP address of the new replacement server.
#On either server1 or server2, do
grep server3 /var/lib/glusterd/peers/*

This will return a uuid followed by ":hostname1=server3"

#On server3, make sure glusterd is stopped, then do
echo UUID={uuid from previous step}>/var/lib/glusterd/glusterd.info

#actual testing below,
[root@node1 ~]# cat /var/lib/glusterd/glusterd.info
UUID=4b9d153c-5958-4dbe-8f91-7b5002882aac
operating-version=30710
#the second line is new.........  maybe not needed...

On server3:
make sure that all brick directories are created/mounted
start glusterd
peer probe one of the existing servers

#restart glusterd, check that full peer list has been populated using
 gluster peer status

(if peers are missing, probe them explicitly, then restart glusterd again)
#check that full volume configuration has been populated using
 gluster volume info

if volume configuration is missing, do 
#on the other node
gluster volume sync "replace-node" all

#on the node to be replaced
setfattr -n trusted.glusterfs.volume-id -v 0x$(grep volume-id /var/lib/glusterd/vols/v1/info | cut -d= -f2 | sed 's/-//g') /gfs/b1/v1
setfattr -n trusted.glusterfs.volume-id -v 0x$(grep volume-id /var/lib/glusterd/vols/v2/info | cut -d= -f2 | sed 's/-//g') /gfs/b2/v2
setfattr -n trusted.glusterfs.volume-id -v 0x$(grep volume-id /var/lib/glusterd/vols/config/info | cut -d= -f2 | sed 's/-//g') /gfs/b1/config/c1

mount -t glusterfs localhost:config /data/data1

#install ctdb if not yet installed and put it back online, use the step on creating the ctdb config but 
#use your common sense not to deleted or modify current one.

gluster vol heal v1 full
gluster vol heal v2 full
gluster vol heal config full

    On Tuesday, August 2, 2016 11:57 AM, Andres E. Moya <amoya@xxxxxxxxxxxxxxxxx> wrote:

 Hi, we have a 2 node replica setup
on 1 of the nodes the file system that had the brick on it failed, not the OS
can we re create a file system and mount the bricks on the same mount point

what will happen, will the data from the other node sync over, or will the failed node wipe out the data on the other mode?

what would be the correct process?

Thanks in advance for any help
_______________________________________________
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