Comment # 10 on bug 1089480 from Wenruo Qu

Btrfs qgroup limit calculation, is based on the principle of, never exceed the
limit even it could return early EDQUOT.

The main challenge for btrfs qgroup is, its reserved metadata space calculation
is pretty unreliable. We over-reserve metadata space even for non-qgroup use
cases.

So current btrfs uses an optimistic method, if we're low on available space, we
try to commit transaction in another thread (commit_kthread) to free some
space.
This behavior is not ensured to free enough space before we hit the limit, so
we could have unexpected EDQUOT related to various factors like system load.

The current behavior almost works fine for x86_64, but since we don't have
ppc64 system to test, I guess it's more obvious to fail such optimistic tries.

Would it be possible to mark that test as soft fail? As we don't have a good
idea to completely fix it soon.