server endpoint to know sync state

[sjpotter]

Currently the restore procedure tells users to do

for node in e{2..3};do
    ssh core@$node "sudo systemctl start etcd2-join.service && sudo systemctl start etcd2.service"
    sleep 10
done

and it sleep to allow the client time to sync, but the note later says to wait until the sync is done and might vary.

could be good to have a etcdctl command that can return the sync status and just loop on that until its ready and then proceed to the next machine.

[xiang90]

There is a metrics in kafka called syncLag to determine the rough sync lag duration between follower and leader. We probably can do similar things for etcd.

[sjpotter]

So one thing I was wondering. when a node joins the cluster is it able to be used by clients before it finishes its initial sync?

[xiang90]

when a node joins the cluster is it able to be used by clients before it finishes its initial sync

It is. But we should probably disable it? See #3879

[sjpotter]

ok, then perhaps as part of that have an endpoint (perhaps /ready) that returns if its ready to be used?

[xiang90]

Now, endpoint status is exposed. User can do endpoint status to check the raft syncing status.