Inconsistent dnames on deployed machines
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
curtin |
Fix Committed
|
Medium
|
Unassigned |
Bug Description
When deploying bcache devices, there is no way to consistently name devices across the machines.
I have three machines, all with identical hardware and with identical configuration (curtin data attached). Device name for / for all machines in MAAS is bcache0. This is also expressed in curtin data.
However, upon deployment, on two machines I get:
/dev/disk/
On third I get:
/dev/disk/
While nobody expect consistent /dev/bcache naming, /dev/disks/
SUBSYSTEM=="block", ACTION=
should in fact be:
SUBSYSTEM=="block", ACTION=
While this solves the problem with filesystems and would uniformly solve /dev/disks/
Can you include the install.log from a system? Which curtin version is being used? The udev rules you show are from curtin 18.1-1 or less. In Bionic and Cosmic (18.1-5) where bcache dnames have been fixed to use uuid of the backing disk look like:
SUBSYSTEM=="block", ACTION= ="add|change" , ENV{CACHED_ UUID}== "272a62c4- ed8f-4714- 8909-78720ffad1 e5", SYMLINK+ ="disk/ by-dname/ cached_ array"