Comment 6 for bug 1619408

I... think i'm ok with unlinking updates once they're a reclaim age old? The risk is dark-data.

The other extreme would introduce like a "check" where if a async update is "old enough" we use direct_client to inspect the available primary object and container nodes and apply some sort of heuristic to decide if the update should be dropped or applied (async PUT and I see the data in majority of primary data nodes but not in any container or async DELETE with 404 or older timestamp on majority of primary data nodes but still in all the container, etc.)

... that's a bunch more code tho - would it be worth it?

I think most of the time I've seen ghost listings come from container replication on re-ingested nodes rather than async's but I don't really know if I can say for sure.

N.B. I don't think the updater is documented to have a reclaim_age option, that said I'm sure everyone has already moved this value to the DEFAULT section - see lp bug #1626296