I was creating 100k items on a fresh cluster and some of couch_db:update_docs calls in mccouch failed thus causing mc_daemon to crash and thus to close a connection. After all sets where done there were 100k items available in memcached. But only about 90k of them were persisted and available from couchdb. Thus it seems that ep-engine or mc-kvstore does not attempt to retry sending commands if the connection to mccouch is being closed.
|For Gerrit Dashboard: &For+MB-4664=message:MB-4664|
|12517,1||MB-4664 Requeue a mutation if a connection to mccouch is reset.||ep-engine||Status: MERGED||+2||+1|