Why it's safe to switch to the new configuration at apply time？

[themanforfree]

Suppose the initial state has three nodes, S1 S2 S3, S1 is current leader.
at t1, and they all use the C1 config.
at t2, user propose a new config C2 to add S4. S1 append C2 to S2 and S3, than leader commit and use C2, S4 start and use C2.
at t3, user propose a new config C3 to add S5. S1 append C3 to S2, S3 and S4, than leader commit and use C3, S4 commit and use C3, S5 start and use C3, S2 and S3 have log entry of C2 C3, but not commited, they always use C1.
In this case, nodes that use C1 and nodes that use C2 have two disjoint majorities, two leaders may be elected. Will this happen in etcd, and if not, what is the reason?

node	t1	t2	t3
S1	C1	C2	C3
S2	C1	C1	C1
S3	C1	C1	C1
S4		C2	C3
S5			C3

[serathius]

Raft supports multiple config changes at once, etcd does not.

[themanforfree]

The above behavior is to make single-node configuration changes twice in a row. After S1 commits C2, S1 can start to process C3, but S2 S3 may not committed C2 yet.