So our staging cluster started going a bit nuts today - began constantly reindexing (blocking other operations. cbcollect_info revealed some map errors but no reduce errors), then when that was resolved (by deleting the views being indexed), it refused to complete a cbbackup.
I was able to determine that the cbbackup tool seemed to be getting stuck consistently after one particular document which, upon closer examination, had an inconsistent revision id. The rev in the doc was at 7 but the rev in the CouchbaseConsole's View editor meta data panel (right hand side) was higher than 500. I found several more docs with such inconsistencies, ranging from a single digit differences to inconsistencies of several thousand.
I tried deleting the offending documents (first via sync gateway delete request followed by python couchbase client for permanent deletion) but upon re-adding the docs I discovered that the meta data panel still somehow saved the old (inconsistent) values for rev_id, despite the docs all being on rev 1 now.
After flushing the bucket and restoring to an older copy of the db, we were able to perform cbbackup again but the inconsistent revs remained. Not sure if related.
Two questions:
1. Why are the revs inconsistent/is this a problem?
2. If it is a problem, is there a way to force consistency?