Re: [PATCH v2 cluster/storage/manager 00/15] storage mapping

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Framework and the iSCSI multi-target rework look like going in the right
direction to me from a high level POV.

The main things that should be fixed:
  - 10/15: the per-mapping ACL path can't be registered and differs from
    the write methods.
  - 09/15: stale-session cleanup can drop a live/last path.

Also: discovery now runs every pvestatd cycle for plain iSCSI storages
(patch 07 & 08) - a behavior change for existing installs. But see the
per-patch replies for more details.

btw. this might be especially nice together with the multipath
improvements I talked about with Friedrich recently - a few things that
might help to prepare for that:

 - Multipath names a LUN by its WWID, but a mapping here is named by
   logical-id + (target, portal). Nothing stores the WWID a mapping ends
   up pointing at, so later there is no easy way to check that all nodes
   see WWID X over K paths. Storing the WWID in the mapping, or at least
   leaving room to look it up when activating, would help.

 - The per-node `targets -> [portals]` that gets resolved is basically
   the "how many paths should we expect here" info such a check needs.
   Right now it only lives inside the ISCSIPlugin $get_local_config
   closure - returning it (or a small helper) would save working it out
   again later.

 - A `multipath` mapping type would not really fit the per-node model: a
   WWID is the same on every node, so it wants one cluster-wide WWID
   list, not one `map` line per node. Not assuming a per-node `node=` is
   always needed in the base plugin would make that easier (see reply on
   02).

Doesn't have to be in your scope now, but I figured I mention it in case
it helps for some synergy or whatnot.