GitLab

Dropping the manager means that setting up node to use disk for given partition becomes manual, but stable across reboots.

Note: /dev/sdN is unstable, so can't be used directly.

There are two actions to be done on machine reboot, after knowing the disk:

• link it to some well place, let it be /dev/disk/disk-path-for-vm (it's not the part of the manager)
• setup the proper permission for that disk (that's the part of the manager)

Those operations for now would have to be done manually.

Example

The operator found out that stable description of a big disk is:

/dev/disk/by-id/ata-SAMSUNG_MZ7LH3T8HMLT-00005_S456NY0M501061

Note: Finding the disk identification is part of another subsystem, which would be a declarative approach to the node.

The operator does somehow the link on the reboot with:

ln -sf $(readlink -f /dev/disk/by-id/ata-SAMSUNG_MZ7LH3T8HMLT-00005_S456NY0M501062) /dev/disk/disk-path-for-vm

And after the instance is allocated there and the operator is aware that the instance can't access the disk, he does the ownership setup on each reboot:

chown slapuser72 $(readlink -f /dev/disk/by-id/ata-SAMSUNG_MZ7LH3T8HMLT-00005_S456NY0M501062)

where the slapuser72 is the allocated instance partition.