swiftgist
commented
6 years ago An additional issue is the engulf functionality. Since this does not detect/import the iSCSI configuration, the default configuration can take over on an upgrade. The "simple" solution may be to disable iSCSI during an engulf. While inconvenient, the admin knows which of the methods above (or something custom) is used.
Martin-Weiss
The issue is the current iSCSI configuration is really an "either or" approach. Lrbd grew independently and preceded DeepSea. Some admins prefer the local editing/managing of the iSCSI configuration since it's centralized in Ceph.
With Salt, configurations are centralized on the master. Generally this is fine and gives the admin the choice of using Jinja templating and possibly automating the management of the configuration. The openATTIC GUI is also relying on this setup.
And there's the crux of the problem: two different locations for the central management. Some admins wish to use all methods (i.e. lrbd -e, Salt and openATTIC) which requires that all accesses write to one location.
To complicate this further, TCMU is on the horizon. The current goal is to support both. I have no idea what this means in terms of seamless management for TCMU.