etcd: panic: tocommit(138789) is out of range [lastIndex(0)]

On 2.2.0 in clustered setup with two static members I’ve initialised second member using ETCD_INITIAL_CLUSTER_STATE="existing" and ETCD_INITIAL_CLUSTER variables. Second member was successfully restarted several times. Then to simulate data loss scenario I’ve stopped second member and deleted its data files. Attempted (re-)start ended up with crash and a long backrace:

2015-09-13 00:45:30.785618 I | rafthttp: the connection with dd3b0cb0cf32bcef became active
2015-09-13 00:45:30.786353 I | raft: c8a988b91c9c29a1 [term: 1] received a MsgApp message with higher term from dd3b0cb0cf32bcef [term: 1676]
2015-09-13 00:45:30.786421 I | raft: c8a988b91c9c29a1 became follower at term 1676
2015-09-13 00:45:30.786491 I | raft: raft.node: c8a988b91c9c29a1 elected leader dd3b0cb0cf32bcef at term 1676
2015-09-13 00:45:30.786509 C | raft: tocommit(138789) is out of range [lastIndex(0)]
panic: tocommit(138789) is out of range [lastIndex(0)]
[...]

I expected second member to be able to recover and re-load data (just like it initialised/loaded its data from first member initially). Even if it such scenario is not intended to work, exit with meaningful error is better than crash. Also I can’t recover from this situation as cluster now is unhealthy and attempt to etcdctl member remove an unavailable (crashing) member fails with Recieved an error trying to remove member c8a988b91c9c29a1: client: etcd cluster is unavailable or misconfigured. It feels fragile as loss of any member (of two) leaves cluster without obvious means to recover or switch to non-clustered (single daemon) setup.