On second thoughts... I think we should just delete this code right now, and stop aging backing files for swap disks. A slightly more conservative option would be to guard the whole lot in a workaround config option which defaults to off.
My reasoning is that the number of different swap backing files is unlikely to be prolific, and is unlikely to grow over time beyond an initial set. We create one of these things for each size of swap file which is specified in a flavor. Firstly, I suspect that this feature isn't widely used at all. But even if it is the total number of swap sizes on offer isn't going to be large: probably less than 10. It's simply not worth this collossal effort to reap them.
Secondly, I may have missed something but I think we're not reaping ephemeral disks. The variety of ephemeral disks is also unlikely to be prolific, but it's going to be more prolific than swap disks, and nobody's screaming about that.
And as I said above, neither of these should have backing files anyway, but that's a harder problem to retrofix.
On second thoughts... I think we should just delete this code right now, and stop aging backing files for swap disks. A slightly more conservative option would be to guard the whole lot in a workaround config option which defaults to off.
My reasoning is that the number of different swap backing files is unlikely to be prolific, and is unlikely to grow over time beyond an initial set. We create one of these things for each size of swap file which is specified in a flavor. Firstly, I suspect that this feature isn't widely used at all. But even if it is the total number of swap sizes on offer isn't going to be large: probably less than 10. It's simply not worth this collossal effort to reap them.
Secondly, I may have missed something but I think we're not reaping ephemeral disks. The variety of ephemeral disks is also unlikely to be prolific, but it's going to be more prolific than swap disks, and nobody's screaming about that.
And as I said above, neither of these should have backing files anyway, but that's a harder problem to retrofix.