I read the document suggest using LVM to make a hu...
# longhorn-storage
c
I read the document suggest using LVM to make a huge volume for all the disk in Longhorn V1。 But if one disk is fail, the LVM volume will be broken. Is it a good practice to use raid to add more redundancy in production, such as make 6 disk in a Raid5/6 group, then use LVM to make a huge volume.
c
I have RAID card in my server. I think it performs better than software RAID. I want to know is it necessary to do a RAID for better protection.
Ceph is not recommended to do RAID.
So what is longhorn?
h
m
Longhorn does replication out of the box to be fault tolerant. Using software raid or hardware raid will hinder performance noticeably fyi.
c
Yes, I understand there is replication。 But imagine A single disk have 99.99% reliability each year. Using 24 disk to make a large LVM volume, the reliability of the volume will become 99.99% power 24, the result is 23%. So a replication reliability is not high.
👍 1
b
We are planning to use LVM across subsets of disks in the physical nodes. e.g. turning 6 x 2TB drives in each node into the equivalent of 3 x 4TB drives so a disk failure won't take down an entire node's storage and longhorn has an easier time with scheduling
if your typical PVC size is significantly smaller than your physical disks there is probably no need
c
@brainy-kilobyte-33711 After last weekend research, I still believe RAID(or even Hardware RAID) is important for Longhorn V1‘s LVM volume in prod. Just think the disk containing your ext4 filesystem inodes is broken, your total filesystem is broken. All your pvc file is lost. I decide to use 5 disks as a group to form a hardware RAID 5. Longhorn V2 engine uses raw disk to manage replications. When its "Single Logical Volume Store Across Multiple Disks" feature is ready, I think it can do without RAID in prod
b
As mentioned before, would longhorn's replication not mitigate that risk? If a disk in a node fails and takes down the entire LVM volume group, your data would still be available on other replicas on other nodes. V2 will hopefully also support data shading https://github.com/longhorn/longhorn/issues/1061
c
yes, but a LVM volume's replication is so easy to be broken. I think you need at least three replications to achieve a pretty good protection. With Raid 5 and 2 replication, it saves your cost.
And longhorn need to rebuild this entire large volume's pvc. It need your to have enough space and also eats your resources.