[gpfsug-discuss] migrate to new metadata GNR systems while maintaining a fallback possibilty

[Billich Heinrich Rainer (ID SD)]

Hello,

 

I need to update/replace our metadata disks but I want to keep old and new in parallel for a while before I remove the old storage: as active/active pair with double copies. This will allow an immediate fall-back if we ever need.

 

Maybe you want to comment on this procedure – I probably found it some time ago on this mailing list, sorry if I don’t remember the author.

 

Some concerns I have

I’m a bit worried what happens when we remove the vdisks with the second copy in the last step – will be get millions of error messages?  Is the sequence of commands right? 

I need to change the failure groups of some vdisks while in use – I wonder if this poses some risk? As I understand this will change the order of block allocation among the nsds (not the allocation map I guess)

This will double metadata write-io, the systems should be able to handle it. 

We will get  better metadata read-io during the transition than what we’ll finally get.

 

== Start – ‘old’ ESS/GNR only

 

 -m 1 (single metadata copy), 

 -M 2

 -K whenpossible

 Metadata vdisks in failure groups 1 and 2

 

== Preparation

 

Use’ mmvdisk filesystems’ and move all metadata vdisk-sets to a single failuregroup (this has to be done in operation, under load)

Set  -m 2, while we still have one failure group only. -K whenpossible will keep us running

Add the new vdisk-set with a second failure group on ‘new’ ESS/GNR systems

Now all new inodes have two copies, one on old and one on new

 

== Create copies on ‘old’ and ‘new’

 

 mmrestripefs -R with –qos maintenance and -N helper-nodes to minimize the load. This may create some locks on the filesystem/cluster and interfere with backup and snapshots??

Maybe better: use a policy ‘replicate’ rule to replicate all files, I can run this in small chunks and run mmrestripefs just once to crosscheck.

 

== Observe for some days, handle remaining filesystems in the same way

 

== Finally

 

Suspend ‘old’ disks

Run mmrestripefs -m

Remove ‘old’ vdisk sets with mmvidisk – this will run another mmdeldisk

Change to -m 1

Run fix replication setting mmrestripefs -R (if needed?)

 

Thank you for reading and for any comments or suggestions.

 

Heiner

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gpfsug.org/pipermail/gpfsug-discuss_gpfsug.org/attachments/20210510/bf8bb6f9/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5254 bytes
Desc: not available
URL: <http://gpfsug.org/pipermail/gpfsug-discuss_gpfsug.org/attachments/20210510/bf8bb6f9/attachment-0001.bin>