This message was deleted.
# longhorn-storage
a
This message was deleted.
f
Hey @chilly-jewelry-23407, glad to hear you're trying out Longhorn! As you discovered, the Longhorn docs don't currently make any particular recommendations concerning hardware RAID vs software RAID vs LVM or particular RAID levels. I'll be interested to see what kind of feedback you get from the community on this one. Here are some things I'd consider: • Longhorn is perfectly capable of making use of a bunch of independent disks. It will schedule replicas to the disks with the most available space and rebuild the data for the appropriate replicas if a disk ever fails and must be permanently removed. In many ways, this is actually the simplest configuration, especially if your volumes are small relative to the size of your physical disks. (If you planned on scheduling 600 GiB volumes to 800 GiB disks, you would, of course, end up losing a decent amount of usable capacity. • I think it's fine to stripe pairs or trios of disks together in RAID 0s, but there is probably a size of storage "pool" that would benefit from some local hardware/software/LVM RAID 5, 6, etc. redundancy. If all of the disks on a node are striped and one disk fails, Longhorn is certainly capable of rebuilding the data in that striped pool from replicas once it's back up again. However, this might require a very large network transfer that would take a lot of time and eat a lot of bandwidth. It may be better to limit the failure domain to a single disk (by just exposing the physical disks to Longhorn) or configuring a self-healing RAID that wouldn't require rebuilding by Longhorn.
🙏 1
c
Thank you @faint-sunset-36608 I really do appreciate your time and help - explaining what I should consider! Semi-related: I watched a Youtube presentation on the new SPDK, and see it's a preview feature. Exciting times! (although this is likely too exciting for me right now)
b
We have a similar setup, and landed on not doing any RAID at all and just exposing the disks to longhorn knowing that the data is duplicated against failure. We actually had a disk go bad recently and were glad we didn't do anything more fancy than just exposing the disks. Longhorn handled it like a pro, so consider this a +1 for that setup.
🙏 1
🚀 1