There's a pre-existing IndexError in the pop from the container node
list in reaper's reap_object method for object rings with a replica
count greater than the container replica count. Which is more likely
on EC storage policies.
When making the backend direct delete requests to the nodes once the
container node's list is exhausted the generic exception handler logs
the error and breaks out of any other backend object requests - but
the reaper marches forward and eventually the tombstones are
replicated.
This change just cycles the container headers across all the nodes -
which seems reasonable enough - but could certainly garner
bikeshedding.
Change-Id: I5897d00b0a8c1e05884945dd93d9ce891b207001
b2189ef47a