In cartridge we've encountered the following case:
We've got two instances, m-m replication is already set up. One is leader (rw), the other one is read-only.
We enable vshard-storage role, and both instances call vshard.storage.cfg almost simultaneously.
If replica calls vshard.storage.cfg before master, it fails.
In v0.1.15 it results in error vshard/storage/init.lua:2389: attempt to index field '_bucket' (a nil value).
In earlier versions it resulted in cfg hanging forever, because of box.once which in turn calls wait_rw which never ends on replica.
In cartridge we've encountered the following case:
We've got two instances, m-m replication is already set up. One is leader (rw), the other one is read-only. We enable vshard-storage role, and both instances call
vshard.storage.cfgalmost simultaneously.If replica calls
vshard.storage.cfgbefore master, it fails.In v0.1.15 it results in error
vshard/storage/init.lua:2389: attempt to index field '_bucket' (a nil value).In earlier versions it resulted in
cfghanging forever, because ofbox.oncewhich in turn callswait_rwwhich never ends on replica.