[ClusterLabs Developers] iSCSILogicalUnit

[Ken Gaillot]

On Sun, 2017-12-17 at 09:15 -0500, Patrick Domack wrote:
> The iSCSILogicalUnit lio-t is not operational with targetcli 3.x
> 
> I have made the following adjustments for a few years now that
> mostly  
> resolved the issue, but failovers where not solid. When looking
> into  
> this again, I noticed the version on github had been updated to  
> include some locking behavior, and when that was added long with
> the  
> changes I made to make it targetcli 3.x compatable, I haven't had
> any  
> issues since.
> 
> Attached is the patch. Mainly I wanted to know the best why you
> would  
> want to commit it. As I believe you wouldn't take it as is, since
> it  

That's great, thanks! If you're comfortable with it, the best way to
submit it is via a github pull request:

https://github.com/ClusterLabs/resource-agents

You would need a github account where you can put up your commits. If
you're not familiar with it, I can give you step-by-step directions.

If you'd rather not, we can put up a pull request with the patch, and
we can discuss it further there.

> removed targetcli <3.x support. Should it detect what version of  
> targetcli is being used on the system? Should it be a new section,  
> like lio-t3, or another method.

I'll let someone more familiar with iSCSI comment about auto-detection, 
but worst case, we can always add a parameter that users can set to
select the targetcli compatibility, or maybe add more flavors to the
"implementation" parameter (lio-t-t3?).

> I also have issues with the new behavior that was added lately to
> it,  
> that adds an ipv6 ip address to the iscsi target group. I don't  
> understand why this code is in the LogicalUnit setup, and not in
> the  
> iSCSITarget code, and also why it's even in there at all. This  
> completely breaks my IP config in my iSCSITarget config, it has no  
> checks to see what was configured but just blanket changes the
> config  
> of the target everytime a lun is added to that target. It only
> checks  
> if ipv6 is active, and if so adds it. It doesn't even bother to
> check  
> if ipv6 is enabled on the target interface or anything, just
> anywhere  
> on the system (localhost, only link scope). This can be done, but  
> should ONLY be done in iSCSITarget, if no binding IP address was  
> defined.

That all sounds reasonable. The developers who work on it should be
able to comment further. FYI, with a github account, you can "watch"
the repository to be notified of proposed changes (pull requests), and
comment on them before they're adopted. People who submit pull requests
and the developers who review them by necessity are primarily familiar
with their own use cases, and it really helps to have other users'
viewpoints.
-- 
Ken Gaillot <kgaillot at redhat.com>