LVM/RAID again
I've been reading several LVM/RAID postings on this mailinglist to find out if someone else is also having the same question as I have, but this does not seem to be the case; I hope this is the right place to ask this.
As LVM does not provide RAID1 (etc.) as VXVM does with its plexes, I wanted to use LVM on top of the standard Linux metadevices (RAID1). However, I wanted to do this for a cluster system using a shared SCSI bus. I am aware of the fact that ownership of disks is important, while something like LVM actually tries to abstract this away, but I want to use LVM to be able to extend the filesystem, to take snapshot backups etc.. The problem I currently face however, is the fact that both nodes autodetect the metadevices and start syncing the mirrors, independently of each other (I do not have the LVM and cluster software installed on these machines yet). The synchronization however, is initiated by the raidstart command, that is true (it is not the detection itself which does this), but I wonder how I should solve this problem. Below, some ideas that come to my mind:
- do not use LVM at all, just use the physical disks instead
- have two separate raidtab files where each of them is only "owned" by a single node (to prevent syncing problems) and only start the raid for a certain metadevice when a logical volume is imported on a node
What do you think about this?
Signature · 1 line
[email protected]
Having used and despised VxVM on Solaris before, and finding out about Heartbeat since then, I'd look into attached enclosures shared physically on a SCSI bus (there is an article in SysAdmin magazine that I used for my starting point) and serial heartbeats across the nodes, if the cluster is simple enough... if not post to the linux-HA list :-)
Quoth Frank Lenaerts:
Quoted 30 lines
I've been reading several LVM/RAID postings on this mailinglist to find out if someone else is also having the same question as I have, but this does not seem to be the case; I hope this is the right place to ask this.
As LVM does not provide RAID1 (etc.) as VXVM does with its plexes, I wanted to use LVM on top of the standard Linux metadevices (RAID1). However, I wanted to do this for a cluster system using a shared SCSI bus. I am aware of the fact that ownership of disks is important, while something like LVM actually tries to abstract this away, but I want to use LVM to be able to extend the filesystem, to take snapshot backups etc.. The problem I currently face however, is the fact that both nodes autodetect the metadevices and start syncing the mirrors, independently of each other (I do not have the LVM and cluster software installed on these machines yet). The synchronization however, is initiated by the raidstart command, that is true (it is not the detection itself which does this), but I wonder how I should solve this problem. Below, some ideas that come to my mind:
- do not use LVM at all, just use the physical disks instead
- have two separate raidtab files where each of them is only "owned" by a single node (to prevent syncing problems) and only start the raid for a certain metadevice when a logical volume is imported on a node
What do you think about this?
Signature · 1 line
[email protected]
Signature · 3 lines
"To be suspended from the legal profession is the moral equivalent of
being ostracized by child molesters."
--Ian RowanQuoted 3 lines · Frank Lenaerts
take snapshot backups etc.. The problem I currently face however, is the fact that both nodes autodetect the metadevices and start syncing the mirrors, independently of each other (I do not have the LVM and
...
Quoted 1 line
- do not use LVM at all, just use the physical disks instead
this is not an lvm problem but an md one
list is [email protected]
Quoted 4 lines
- have two separate raidtab files where each of them is only "owned" by a single node (to prevent syncing problems) and only start the raid for a certain metadevice when a logical volume is imported on a node
get rid of raidstart and raidtab and use
mdctl from neil brown
http://www.cse.unsw.edu.au/~neilb/source/mdctl/
also be sure you are not using raid autodetection
L.
Signature · 6 lines
Luca Berra -- [email protected] Communication Media & Services S.r.l. /"\ \ / ASCII RIBBON CAMPAIGN X AGAINST HTML MAIL / \