Question about voting restarted server as leader

[zhaoguo wang]

Hi, 

I am new in this group. I am trying to study Raft in etcd and have a question about the voting implementation.

In the current implementation, after a server is restarted, it can not be elected to be leader anymore as it is not "promotable". 

My question is what is should be done to make a restart machine to be promotable? 

By checking the source code, I found it may be solved by creating a snapshot which will restore the peers' progress. 

But what if there is no snapshot has been created before?

The following test will get stuck as no server can be voted as a leader at the end. (I run this test by adding it to /coreos/etcd/raft/rafttest/node_test.go)

func Test(t *testing.T) {

peers := []raft.Peer{{1, nil}, {2, nil}, {3, nil}}

nt := newRaftNetwork(1, 2, 3)

nodes := make([]*node, 0)

for i := 1; i <= 3; i++ {

n := startNode(uint64(i), peers, nt.nodeNetwork(uint64(i)))

nodes = append(nodes, n)

}

waitLeader(nodes)

nodes[0].stop()

time.Sleep(time.Millisecond)

nodes[0].restart()

if nodes[0].Status().SoftState.Lead == 0 {

time.Sleep(time.Millisecond)

}

nodes[1].stop()

time.Sleep(time.Millisecond)

nodes[1].restart()

if nodes[1].Status().SoftState.Lead == 0 {

time.Sleep(time.Millisecond)

}

nodes[2].stop()

time.Sleep(time.Millisecond)

nodes[2].restart()

if nodes[2].Status().SoftState.Lead == 0 {

time.Sleep(time.Millisecond)

}

        nodes[0].Propose(context.TODO(), []byte("somedata"))

        if !waitCommitConverge(nodes, 1) {

t.Errorf("commits failed to converge!")

}

for _, n := range nodes {

n.stop()

}

}

Thank you so much

Best,

Zhaoguo