[RFE] Support for Loopback devices as bricks

[aravindavk]

Auto provisioning Volumes by default provisions the bricks using LVM. Alternatively Loop devices can be used to create Bricks.

Register device(LVM vs Loop):

LVM	Loop
Create PV	-
Create VG	-
Collect Available size and save details in Store	Collect Available size and save details in Store

Unregister device

LVM	Loop
Remove VG	-
Remove PV	-
Remove details from Store if no bricks exists	Remove details from Store if no bricks exists

Create Brick

LVM	Loop
Create Thinpool	Create directory
Create LV	Create Loop device
Create FS	Create FS
Mount LV	Mount device
Create Brickdir	Create Brickdir

Mount/Unmount Brick

LVM	Loop
Mount/Unmount LV	Mount/Unmount loop device

Remove Brick

LVM	Loop
Unmount brick if mounted	Unmount brick if mounted
Remove LV	Remove loop file
Remove Thinpool if number of LVs is zero	Remove directory if no files exists

Get Available Size

LVM	Loop
Using LVM commands	Using statvfs or df command

[JohnStrunk]

For "loop", do we have an opinion on the device being given to gd2? Should it be a block device or a mounted file system? (I ask because there's no mention of formatting when the device is registered)

[aravindavk]

For "loop", do we have an opinion on the device being given to gd2? Should it be a block device or a mounted file system? (I ask because there's no mention of formatting when the device is registered)

I was thinking directory path in mounted file system. Do we need to manage the bricks hosting Volume?

[JohnStrunk]

We'll already have sufficient permissions, so we could format and mount the lower fs ourselves. It also provides us the opportunity to add vdo in the future.

We'd need to examine whether there are reliability implications though since we would be responsible for the unmount at shutdown.

I don't know which way we should go with this. Probably start with the file system just because there are fewer unknowns there.

[poornimag]

With LVM being replaced by loop devices, How do we plan to handle disks with different size,with LVM the disks are pooled and carved to required sizes? Since the loopback brick cannot span across devices? The +1 scale will become more restrictive? Users will have to add disks of same/minimum size devices?

[JohnStrunk]

The brick can be created with any size up to the size of the lower-level file system. The restriction this is adding is that we can't string together multiple physical devices (lower-level file systems) and create a single brick that spans that whole capacity. We do have the ability to use dht to glue the smaller bricks together.
Going forward, we actually want to limit the max size of a brick to ease load and capacity balancing, so this may not be a big issue in practice.