zFCP improvements

[joseivanlopez]

Problem

Two things to fix/improve:

• Automatically activated zFCP disks are not probed after activating a controller.
• There is no info about allow_lun_scan when the option is disabled.

Solution

• Delay zFCP probing in order to give some time to disk activation.
• The system is marked as deprecated after probing, if needed.
• allow_lun_scan status is always informed.
• The allow_lun_scan explanation is improved.

Bonus: members of a device are sorted in the disk selector.

Testing

• Adapted unit tests
• Tested manually

Screenshots

Show/hide

[coveralls]

coverage: 72.531% (+0.03%) from 72.505% when pulling 967e495 on joseivanlopez:zfcp-improvements into f6152b0 on openSUSE:master.

[imobachgs]

Just some minor comments on the changes file format.

About the fixes, they look good. We might need to find a better way than a call to sleep, but we do not have any now.

What does it happen if it takes more than 2 seconds? Would the user eventually see the list of disks? And, in that case, can the user select one of them for installation?

[joseivanlopez]

What does it happen if it takes more than 2 seconds? Would the user eventually see the list of disks? And, in that case, can the user select one of them for installation?

According to our tests, 2 seconds should be more than enough. Anyway, if exporting disks takes longer and the disks are not read, then zFCP has to be probed again. Right now there is nothing in the UI for doing the reprobing. To force a zFCP probing you have to leave the zFCP page (going back) and visit it again. The system is properly marked as deprecated (if needed) once the list of zFCP disks is read.

As said, this should not happen with the sleep delay. And I would avoid adding a button for manually re-read zFCP only just in case. What do you think?

Update: After discussing it, this issue should be solved in a more generic way, see #652.

[imobachgs]

np: Is it possible that @on_disks_change_callbacks is nil? Not something to change in this PR, but I rather prefer to have an empty array.

[joseivanlopez]

Yes, because it is initialize on assignement, see https://github.com/openSUSE/agama/blob/master/service/lib/agama/storage/zfcp/manager.rb#L51. We could avoid it by initializing the list in the constructor or by adding a #on_disk_change_callbacks method.