Reconfigure Gluster from Replica 3 to Arbiter 1/Replica 2
My current hyperconverged environment is replicating data across all 3 servers. I'm running critically low on disk space, and need to add space. To that end, I've ordered 8x 800GB ssd drives, and plan to put 4 drives in 1 server, and 4 drives in the other. What's my best option for reconfiguring the hyperconverged cluster, to change gluster storage away from Replica 3 to a Replica 2 / Arbiter model? I'd really prefer not to have to reinstall things from scratch, but I'll do that if I have to. My most important requirement is that I cannot have any downtime for my VMs (so I can only reconfigure 1 host at a time). Sent with ProtonMail Secure Email.
Hi David, any storage operation can cause unexpected situations, so always plan your activites for low traffic hours and test them on your test environment in advance. I think it's easier if you (command line): -verify no heals are pending. Not a single one.- set the host to maintenance over ovirt- remove the third node from gluster volumes (remove-brick replica 2)- umount the bricks on the third node- recreate a smaller LV with '-i maxpct=90 size=512' and mount it with the same options like the rest of the nodes. Usually I use 'noatime,inode64,context=system_u:object_r:glusterd_brick_t:s0'- add this new brick (add-brick replica 3 arbiter 1) to the volume- wait for the heals to finish Then repeat again for each volume. Adding the new disks should be done later. Best Regards,Strahil Nikolov On Sat, Jul 10, 2021 at 3:15, David White via Users<users@ovirt.org> wrote: My current hyperconverged environment is replicating data across all 3 servers. I'm running critically low on disk space, and need to add space. To that end, I've ordered 8x 800GB ssd drives, and plan to put 4 drives in 1 server, and 4 drives in the other. What's my best option for reconfiguring the hyperconverged cluster, to change gluster storage away from Replica 3 to a Replica 2 / Arbiter model? I'd really prefer not to have to reinstall things from scratch, but I'll do that if I have to. My most important requirement is that I cannot have any downtime for my VMs (so I can only reconfigure 1 host at a time). Sent with ProtonMail Secure Email. _______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/YHEAXRUNA2RDJR...
Thank you. And yes, I agree, this needs to occur in a maintenance window and be done very carefully. :) My only problem with this method is that I need to *replace* disks in the two servers. I don't have any empty hard drive bays, so will effectively need to put a host into maintenance mode, remove the drives, and put new drives in. I will NOT be touching the OS drives, however, as those are on their own separate RAID array. So, essentially, it will need to look something like this: - Put the cluster into global maintenance mode - Put 1 host into full maintenance mode / deactivate it - Stop gluster - Remove the storage - Add the new storage & reconfigure - Start gluster - Re-add the host to the cluster Adding the new storage & reconfiguring is the head scratcher for me, given that I don't have room for the old hard drives + new hard drives at the same time. Sent with ProtonMail Secure Email. ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ On Saturday, July 10th, 2021 at 5:55 AM, Strahil Nikolov <hunter86_bg@yahoo.com> wrote:
Hi David,
any storage operation can cause unexpected situations, so always plan your activites for low traffic hours and test them on your test environment in advance.
I think it's easier if you (command line):
-verify no heals are pending. Not a single one.- set the host to maintenance over ovirt- remove the third node from gluster volumes (remove-brick replica 2)- umount the bricks on the third node- recreate a smaller LV with '-i maxpct=90 size=512' and mount it with the same options like the rest of the nodes. Usually I use 'noatime,inode64,context=system_u:object_r:glusterd_brick_t:s0'- add this new brick (add-brick replica 3 arbiter 1) to the volume- wait for the heals to finish
Then repeat again for each volume.
Adding the new disks should be done later.
Best Regards,Strahil Nikolov
On Sat, Jul 10, 2021 at 3:15, David White via Users<users@ovirt.org> wrote:My current hyperconverged environment is replicating data across all 3 servers. I'm running critically low on disk space, and need to add space.
To that end, I've ordered 8x 800GB ssd drives, and plan to put 4 drives in 1 server, and 4 drives in the other.
What's my best option for reconfiguring the hyperconverged cluster, to change gluster storage away from Replica 3 to a Replica 2 / Arbiter model? I'd really prefer not to have to reinstall things from scratch, but I'll do that if I have to.
My most important requirement is that I cannot have any downtime for my VMs (so I can only reconfigure 1 host at a time).
Sent with ProtonMail Secure Email.
_______________________________________________
Users mailing list -- users@ovirt.org
To unsubscribe send an email to users-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/YHEAXRUNA2RDJR...
Hmm.... right as I said that, I just had a thought. I DO have a "backup" server in place (that I haven't even started using yet), that currently has some empty hard drive bays. It would take some extra work, but I could use that 4th backup server as a temporary staging ground to begin building the new Gluster configuration. Once I have that server + 2 of my production servers rebuilt properly, I could then simply remove and replace this "backup" server with my 3rd server in the cluster. So this effectively means that I have 2 servers that I can take down completely at a single time to rebuild gluster, instead of just 1. I think that simplifies things. Sent with ProtonMail Secure Email. ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ On Saturday, July 10th, 2021 at 6:14 AM, David White <dmwhite823@protonmail.com> wrote:
Thank you. And yes, I agree, this needs to occur in a maintenance window and be done very carefully. :)
My only problem with this method is that I need to *replace* disks in the two servers. I don't have any empty hard drive bays, so will effectively need to put a host into maintenance mode, remove the drives, and put new drives in.
I will NOT be touching the OS drives, however, as those are on their own separate RAID array.
So, essentially, it will need to look something like this:
- Put the cluster into global maintenance mode - Put 1 host into full maintenance mode / deactivate it - Stop gluster - Remove the storage - Add the new storage & reconfigure - Start gluster - Re-add the host to the cluster
Adding the new storage & reconfiguring is the head scratcher for me, given that I don't have room for the old hard drives + new hard drives at the same time.
Sent with ProtonMail Secure Email.
‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Saturday, July 10th, 2021 at 5:55 AM, Strahil Nikolov <hunter86_bg@yahoo.com> wrote:
Hi David,
any storage operation can cause unexpected situations, so always plan your activites for low traffic hours and test them on your test environment in advance.
I think it's easier if you (command line):
-verify no heals are pending. Not a single one.- set the host to maintenance over ovirt- remove the third node from gluster volumes (remove-brick replica 2)- umount the bricks on the third node- recreate a smaller LV with '-i maxpct=90 size=512' and mount it with the same options like the rest of the nodes. Usually I use 'noatime,inode64,context=system_u:object_r:glusterd_brick_t:s0'- add this new brick (add-brick replica 3 arbiter 1) to the volume- wait for the heals to finish
Then repeat again for each volume.
Adding the new disks should be done later.
Best Regards,Strahil Nikolov
On Sat, Jul 10, 2021 at 3:15, David White via Users<users@ovirt.org> wrote:My current hyperconverged environment is replicating data across all 3 servers. I'm running critically low on disk space, and need to add space.
To that end, I've ordered 8x 800GB ssd drives, and plan to put 4 drives in 1 server, and 4 drives in the other.
What's my best option for reconfiguring the hyperconverged cluster, to change gluster storage away from Replica 3 to a Replica 2 / Arbiter model? I'd really prefer not to have to reinstall things from scratch, but I'll do that if I have to.
My most important requirement is that I cannot have any downtime for my VMs (so I can only reconfigure 1 host at a time).
Sent with ProtonMail Secure Email.
_______________________________________________
Users mailing list -- users@ovirt.org
To unsubscribe send an email to users-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/YHEAXRUNA2RDJR...
Just a thought but depending on resources you might be able to use your 4th server as nfs storage and live migrate vm disks to it and off of your gluster volumes. I’ve done this in the past when doing major maintenance on gluster volumes to err on the side of caution. On Sat, Jul 10, 2021 at 7:22 AM David White via Users <users@ovirt.org> wrote:
Hmm.... right as I said that, I just had a thought. I DO have a "backup" server in place (that I haven't even started using yet), that currently has some empty hard drive bays.
It would take some extra work, but I could use that 4th backup server as a temporary staging ground to begin building the new Gluster configuration. Once I have that server + 2 of my production servers rebuilt properly, I could then simply remove and replace this "backup" server with my 3rd server in the cluster.
So this effectively means that I have 2 servers that I can take down completely at a single time to rebuild gluster, instead of just 1. I think that simplifies things.
Sent with ProtonMail <https://protonmail.com/> Secure Email.
‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Saturday, July 10th, 2021 at 6:14 AM, David White < dmwhite823@protonmail.com> wrote:
Thank you. And yes, I agree, this needs to occur in a maintenance window and be done very carefully. :)
My only problem with this method is that I need to *replace* disks in the two servers. I don't have any empty hard drive bays, so will effectively need to put a host into maintenance mode, remove the drives, and put new drives in.
I will NOT be touching the OS drives, however, as those are on their own separate RAID array.
So, essentially, it will need to look something like this:
- Put the cluster into global maintenance mode - Put 1 host into full maintenance mode / deactivate it - Stop gluster - Remove the storage - Add the new storage & reconfigure - Start gluster - Re-add the host to the cluster
Adding the new storage & reconfiguring is the head scratcher for me, given that I don't have room for the old hard drives + new hard drives at the same time.
Sent with ProtonMail <https://protonmail.com/> Secure Email.
‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ On Saturday, July 10th, 2021 at 5:55 AM, Strahil Nikolov < hunter86_bg@yahoo.com> wrote:
Hi David,
any storage operation can cause unexpected situations, so always plan your activites for low traffic hours and test them on your test environment in advance.
I think it's easier if you (command line):
-verify no heals are pending. Not a single one. - set the host to maintenance over ovirt - remove the third node from gluster volumes (remove-brick replica 2) - umount the bricks on the third node - recreate a smaller LV with '-i maxpct=90 size=512' and mount it with the same options like the rest of the nodes. Usually I use 'noatime,inode64,context=system_u:object_r:glusterd_brick_t:s0' - add this new brick (add-brick replica 3 arbiter 1) to the volume - wait for the heals to finish
Then repeat again for each volume.
Adding the new disks should be done later.
Best Regards, Strahil Nikolov
On Sat, Jul 10, 2021 at 3:15, David White via Users <users@ovirt.org> wrote: My current hyperconverged environment is replicating data across all 3 servers. I'm running critically low on disk space, and need to add space.
To that end, I've ordered 8x 800GB ssd drives, and plan to put 4 drives in 1 server, and 4 drives in the other.
What's my best option for reconfiguring the hyperconverged cluster, to change gluster storage away from Replica 3 to a Replica 2 / Arbiter model? I'd really prefer not to have to reinstall things from scratch, but I'll do that if I have to.
My most important requirement is that I cannot have any downtime for my VMs (so I can only reconfigure 1 host at a time).
Sent with ProtonMail <https://protonmail.com/> Secure Email.
_______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/YHEAXRUNA2RDJR...
_______________________________________________ Users mailing list -- users@ovirt.org To unsubscribe send an email to users-leave@ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/WWETWL7N7FT5UC...
HI Jayme & Strahil, Thank you again for your messages. Reading https://stackoverflow.com/questions/52394849/can-i-change-glusterfs-replica-..., I think I understand now what Strahil is suggesting. It sounds to me like you're saying I can reconfigure the existing gluster volume (don't destroy it) by changing it from a Replica 3 over to a Replica 2/Arbiter 1. Then once I do that operation, I can THEN go in and put the two hosts into maintenance mode one at a time to rebuild the RAID arrays, and then heal the volume on the larger RAID. So it would look like this: 1) Put 3rd host into maintenance mode & verify no heals are necessary 2) Make the 3rd host the arbiter 2a) remove-brick replica 2 -- if I understand correctly, this will basically just reconfigure the existing volume to replicate between the 2 bricks, and not all 3 ... is this correct? 2b) add-brick replicat 3 arbiter 1 -- If I understand correctly, this will reconfigure the volume (again), adding the 3rd server's storage back to the Gluster volume, but only as an arbiter node, correct? 3) Now with everything healthy, the volume is now a Replica 2 / Arbiter 1.... and I can now stop gluster on each of the 2 servers getting the storage upgrade, rebuild the RAID on the new storage, reboot, and let gluster heal itself before moving on to the next server. Do I understand this process right? Jayme's idea to use my 4th server as a temporary NFS location is not a bad idea. I could definitely do that, "just in case" the Gluster volume got corrupted. Unfortunately my backup server has spinning disks instead of SSD drives, but speed wouldn't be too noticeable, and I could do it over a weekend or something. Thanks again for your input. Sent with ProtonMail Secure Email. ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ On Saturday, July 10th, 2021 at 6:53 AM, Jayme <jaymef@gmail.com> wrote:
Just a thought but depending on resources you might be able to use your 4th server as nfs storage and live migrate vm disks to it and off of your gluster volumes. I’ve done this in the past when doing major maintenance on gluster volumes to err on the side of caution.
On Sat, Jul 10, 2021 at 7:22 AM David White via Users <users@ovirt.org> wrote:
Hmm.... right as I said that, I just had a thought. I DO have a "backup" server in place (that I haven't even started using yet), that currently has some empty hard drive bays.
It would take some extra work, but I could use that 4th backup server as a temporary staging ground to begin building the new Gluster configuration. Once I have that server + 2 of my production servers rebuilt properly, I could then simply remove and replace this "backup" server with my 3rd server in the cluster.
So this effectively means that I have 2 servers that I can take down completely at a single time to rebuild gluster, instead of just 1. I think that simplifies things.
Sent with ProtonMail Secure Email.
‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Saturday, July 10th, 2021 at 6:14 AM, David White <dmwhite823@protonmail.com> wrote:
Thank you. And yes, I agree, this needs to occur in a maintenance window and be done very carefully. :)
My only problem with this method is that I need to *replace* disks in the two servers. I don't have any empty hard drive bays, so will effectively need to put a host into maintenance mode, remove the drives, and put new drives in.
I will NOT be touching the OS drives, however, as those are on their own separate RAID array.
So, essentially, it will need to look something like this:
- Put the cluster into global maintenance mode - Put 1 host into full maintenance mode / deactivate it - Stop gluster - Remove the storage - Add the new storage & reconfigure - Start gluster - Re-add the host to the cluster
Adding the new storage & reconfiguring is the head scratcher for me, given that I don't have room for the old hard drives + new hard drives at the same time.
Sent with ProtonMail Secure Email.
‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Saturday, July 10th, 2021 at 5:55 AM, Strahil Nikolov <hunter86_bg@yahoo.com> wrote:
Hi David,
any storage operation can cause unexpected situations, so always plan your activites for low traffic hours and test them on your test environment in advance.
I think it's easier if you (command line):
-verify no heals are pending. Not a single one.- set the host to maintenance over ovirt- remove the third node from gluster volumes (remove-brick replica 2)- umount the bricks on the third node- recreate a smaller LV with '-i maxpct=90 size=512' and mount it with the same options like the rest of the nodes. Usually I use 'noatime,inode64,context=system_u:object_r:glusterd_brick_t:s0'- add this new brick (add-brick replica 3 arbiter 1) to the volume- wait for the heals to finish
Then repeat again for each volume.
Adding the new disks should be done later.
Best Regards,Strahil Nikolov
On Sat, Jul 10, 2021 at 3:15, David White via Users<users@ovirt.org> wrote:My current hyperconverged environment is replicating data across all 3 servers. I'm running critically low on disk space, and need to add space.
To that end, I've ordered 8x 800GB ssd drives, and plan to put 4 drives in 1 server, and 4 drives in the other.
What's my best option for reconfiguring the hyperconverged cluster, to change gluster storage away from Replica 3 to a Replica 2 / Arbiter model? I'd really prefer not to have to reinstall things from scratch, but I'll do that if I have to.
My most important requirement is that I cannot have any downtime for my VMs (so I can only reconfigure 1 host at a time).
Sent with ProtonMail Secure Email.
_______________________________________________
Users mailing list -- users@ovirt.org
To unsubscribe send an email to users-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/YHEAXRUNA2RDJR...
_______________________________________________
Users mailing list -- users@ovirt.org
To unsubscribe send an email to users-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/WWETWL7N7FT5UC...
2a) remove-brick replica 2 -- if I understand correctly, this will basically just reconfigure the existing volume to replicate between the 2 bricks, and not all 3 ... is this correct?
Yep, you are kicking out the 3rd node and the volume is converted to replica 2. Most probably the command would be gluster volume remove-brick vmstore replica 2 host3:/path/to/brick force
2b) add-brick replicat 3 arbiter 1 -- If I understand correctly, this will reconfigure the volume (again), adding the 3rd server's storage back to the Gluster volume, but only as an arbiter node, correct? Yes,I would prefer to create a fresh new LV.Don't forget to raise the inode count higher,as this one will be an arbiter brick (see previous e-mail). Once you add via gluster volume add-brick vmstore replica 3 arbiter 1 host3:/path/to/new/brick , you will have to wait for all heals to complete
3) Now with everything healthy, the volume is now a Replica 2 / Arbiter 1.... and I can now stop gluster on each of the 2 servers getting the storage upgrade, rebuild the RAID on the new storage, reboot, and let gluster heal itself before moving on to the next server.
If you rebuild the raid, you are destroying the brick, so after mounting it back, you will need to reset-brick. If it doesn't work for some reason , you can always remove-brick replica 1 host1:/path/to/brick arbiter:/path/to/brick and readd them with add-brick replica 3 arbiter 1. I had some paused VMs after raid reshaping (spinning disks) during the healing but my lab is running on workstations, so do it in the least busiest hours and possible backups should have completed before the reconfiguration and not exactly during the healing ;) Best Regards, Strahil Nikolov
Thank you. I'm doing some more research & reading on this to make sure I understand everything before I do this work. You wrote:
If you rebuild the raid, you are destroying the brick, so after mounting it back, you will need to reset-brick. If it doesn't work for some reason , you can always remove-brick replica 1 host1:/path/to/brick arbiter:/path/to/brick and readd them with add-brick replica 3 arbiter 1.
Would it be safer to remove-brick replica 1 before I destroy the brick? Also, you suggested creating a fresh logical volume on the node where I'm converting from full replica to arbiter. Would it not suffice to simply go in and erase all of the data? I'm still not clear on how to force gluster to use a specific server as the arbiter node. Will gluster just "figure it out" if the logical volume on the arbiter is smaller than on the other two nodes? I'm also still a little bit unclear on why I need to specifically increase the inode when I go to add-brick on the arbiter node. Or is that when (if?) I rebuild the Logical Volume? Do I need to increase the inode on the other two servers after I grow the RAID on the two primary storage servers? Below is an overview of the specific steps I think I need to take, in order: Prepare: Put cluster into global maintenance mode Run the following commands: # gluster volume status data inode host 3 (Arbiter): # gluster volume remove-brick data replica 2 host3.mgt.example.com:/gluster_bricks/data/data # rm -rf /gluster_bricks/data/data/* # ALTERNATIVELY... rebuild the logical volume? Is this really necessary? # gluster volume add-brick data replica 3 arbiter 1 host3.mgt.example.com:/gluster_bricks/data/data (Let the volumes heal completely) host 1 # gluster volume remove-brick data replica 1 host1.mgt.example.com:/gluster_bricks/data/data (rebuild the array & reboot the server) # gluster volume add-brick data replica 3 arbiter 1 host1.mgt.example.com:/gluster_bricks/data/data (Let the volumes heal completely) host 2 # gluster volume remove-brick data replica 1 host1.mgt.example.com:/gluster_bricks/data/data (rebuild the array & reboot the server) # gluster volume add-brick data replica 3 arbiter 1 host1.mgt.example.com:/gluster_bricks/data/data Sent with ProtonMail Secure Email. ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ On Sunday, July 11th, 2021 at 3:14 AM, Strahil Nikolov via Users <users@ovirt.org> wrote:
2a) remove-brick replica 2
-- if I understand correctly, this will basically just reconfigure the existing volume to replicate between the 2 bricks, and not all 3 ... is this correct?
Yep, you are kicking out the 3rd node and the volume is converted to replica 2.
Most probably the command would be gluster volume remove-brick vmstore replica 2 host3:/path/to/brick force
2b) add-brick replicat 3 arbiter 1
-- If I understand correctly, this will reconfigure the volume (again), adding the 3rd server's storage back to the Gluster volume, but only as an arbiter node, correct?
Yes,I would prefer to create a fresh new LV.Don't forget to raise the inode count higher,as this one will be an arbiter brick (see previous e-mail).
Once you add via gluster volume add-brick vmstore replica 3 arbiter 1 host3:/path/to/new/brick , you will have to wait for all heals to complete
3. Now with everything healthy, the volume is now a Replica 2 / Arbiter 1.... and I can now stop gluster on each of the 2 servers getting the storage upgrade, rebuild the RAID on the new storage, reboot, and let gluster heal itself before moving on to the next server.
If you rebuild the raid, you are destroying the brick, so after mounting it back, you will need to reset-brick. If it doesn't work for some reason , you can always remove-brick replica 1 host1:/path/to/brick arbiter:/path/to/brick and readd them with add-brick replica 3 arbiter 1.
I had some paused VMs after raid reshaping (spinning disks) during the healing but my lab is running on workstations, so do it in the least busiest hours and possible backups should have completed before the reconfiguration and not exactly during the healing ;)
Best Regards,
Strahil Nikolov
Users mailing list -- users@ovirt.org
To unsubscribe send an email to users-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/ECHZQWBMC5H3CD...
remove-brick replica 1 will deconfigure your Arbiter which leads to brick prep work on the Arbiter - just more efforts.
Also, you suggested creating a fresh logical volume on the node where I'm converting from full replica to arbiter.Would it not suffice to simply go in and erase all of the data? I can't catch the context properly, but keep in mind that gluster uses extended fs attributes which means that either you have to clean them up, or just 'mkfs xfs -i 512' the whole LV which for me is far simpler. As you will reuse the disks in the other nodes - you most probably will use a new device. Gluster arbiter is always the 3rd entry in the subvolume (3,6,9,etc). By default inodes are a percentage of the free space, and on small disks (you don't want 10TB raid for an arbiter brick, right?) the ammount of inodes will be smaller on the arbiter brick.Example (numbers are made up):Host1 - 10TB disk - 10000 inodesHost2 - 10TB disk - 10000 inodesArbiter - 100G disk - 100 inodes Once you hit the Arbiter inode limit, you can't use it any more. Reusing old bricks is not nice and will prevent you from future nightmares.Just remove the 3rd brick, add new disk on 3rd host, create it fresh , new arbiter, stop gluster. on 1 host, rebuild raid, reset-brick, heal, the other host too.
Best Regards,Strahil Nikolov On Tue, Jul 20, 2021 at 2:48, David White via Users<users@ovirt.org> wrote: Thank you. I'm doing some more research & reading on this to make sure I understand everything before I do this work. You wrote:
If you rebuild the raid, you are destroying the brick, so after mounting it back, you will need to reset-brick. If it doesn't work for some reason , you can always remove-brick replica 1 host1:/path/to/brick arbiter:/path/to/brick and readd them with add-brick replica 3 arbiter 1.
Would it be safer to remove-brick replica 1 before I destroy the brick? Also, you suggested creating a fresh logical volume on the node where I'm converting from full replica to arbiter. Would it not suffice to simply go in and erase all of the data? I'm still not clear on how to force gluster to use a specific server as the arbiter node. Will gluster just "figure it out" if the logical volume on the arbiter is smaller than on the other two nodes? I'm also still a little bit unclear on why I need to specifically increase the inode when I go to add-brick on the arbiter node. Or is that when (if?) I rebuild the Logical Volume? Do I need to increase the inode on the other two servers after I grow the RAID on the two primary storage servers? Below is an overview of the specific steps I think I need to take, in order: Prepare: Put cluster into global maintenance mode Run the following commands: # gluster volume status data inode host 3 (Arbiter): # gluster volume remove-brick data replica 2 host3.mgt.example.com:/gluster_bricks/data/data # rm -rf /gluster_bricks/data/data/* # ALTERNATIVELY... rebuild the logical volume? Is this really necessary? # gluster volume add-brick data replica 3 arbiter 1 host3.mgt.example.com:/gluster_bricks/data/data (Let the volumes heal completely) host 1 # gluster volume remove-brick data replica 1 host1.mgt.example.com:/gluster_bricks/data/data (rebuild the array & reboot the server) # gluster volume add-brick data replica 3 arbiter 1 host1.mgt.example.com:/gluster_bricks/data/data (Let the volumes heal completely) host 2 # gluster volume remove-brick data replica 1 host1.mgt.example.com:/gluster_bricks/data/data (rebuild the array & reboot the server) # gluster volume add-brick data replica 3 arbiter 1 host1.mgt.example.com:/gluster_bricks/data/data Sent with ProtonMail Secure Email. ‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐ On Sunday, July 11th, 2021 at 3:14 AM, Strahil Nikolov via Users <users@ovirt.org> wrote:
2a) remove-brick replica 2
-- if I understand correctly, this will basically just reconfigure the existing volume to replicate between the 2 bricks, and not all 3 ... is this correct?
Yep, you are kicking out the 3rd node and the volume is converted to replica 2.
Most probably the command would be gluster volume remove-brick vmstore replica 2 host3:/path/to/brick force
2b) add-brick replicat 3 arbiter 1
-- If I understand correctly, this will reconfigure the volume (again), adding the 3rd server's storage back to the Gluster volume, but only as an arbiter node, correct?
Yes,I would prefer to create a fresh new LV.Don't forget to raise the inode count higher,as this one will be an arbiter brick (see previous e-mail).
Once you add via gluster volume add-brick vmstore replica 3 arbiter 1 host3:/path/to/new/brick , you will have to wait for all heals to complete
3. Now with everything healthy, the volume is now a Replica 2 / Arbiter 1.... and I can now stop gluster on each of the 2 servers getting the storage upgrade, rebuild the RAID on the new storage, reboot, and let gluster heal itself before moving on to the next server.
If you rebuild the raid, you are destroying the brick, so after mounting it back, you will need to reset-brick. If it doesn't work for some reason , you can always remove-brick replica 1 host1:/path/to/brick arbiter:/path/to/brick and readd them with add-brick replica 3 arbiter 1.
I had some paused VMs after raid reshaping (spinning disks) during the healing but my lab is running on workstations, so do it in the least busiest hours and possible backups should have completed before the reconfiguration and not exactly during the healing ;)
Best Regards,
Strahil Nikolov
Users mailing list -- users@ovirt.org
To unsubscribe send an email to users-leave@ovirt.org
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/
List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/ECHZQWBMC5H3CD...
participants (3)
-
David White -
Jayme -
Strahil Nikolov