Fix etcd v3.4 cluster works

[ainoniwa]

This is fix to #127

In etcd v3.4 cluster, required to wait for leader election before serve client.
So, all etcd process need to exec with wait_for_line="listening for peers on" instead wait_for_line="ready to serve client requests".

Additional:

• etcdctl use etcd v3 API (ETCDCTL_API=3) to compatible v3.0~.
• travis to use etcd v3.4.13
• Modify wait_for_line "ready to serve client requests" instead "listening for client requests on" when single process. Probably, the message is already removed. https://github.com/etcd-io/etcd/pull/11592/files#diff-22504f5e6a5f5ff5f4929b7fc571a26735f1b7746f5f143891af245e85ec4f0bL144

[jd]

I guess this is going to break older etcd versions. Is this fine?

[ainoniwa]

Thank you for review it.
This PR will compatible etcd v3.0 or higher. I tested v3.2.31/v3.3.25/v3.4.13, and each etcd UT passed.

In etcd v3.4, ETCDCTL_API=3 is default. so use etcdctl endpoint health is work.
In etcd v3.3 or older, ETCDCTL_API=2 is default. but v3 API is already. so it can use etcdctl cluster-health or export ETCDCTL_API=3; etcdctl endpoint health.

Or, do you mind about other parts?

[jd]

Ok, that sounds great! Thank you!

[felixonmars]

Confirmed that it works with etcd 3.4.15 here. Thanks!