reset watch's lastIndex on error

[alicebob]

Watch()ing a agent in -dev mode misses changes when the agent restarts.

When a -dev agent is restarted it'll have a clean state, including a
reset index. A watch() will reconnect after a restart of the agent, but it won't
notice that the index counter has reset and it will keep waiting until
the server reached the old index again, which is wrong. Resetting the index
prevents that and makes watch() work for -dev agents over restarts.

[slackpad]

Hi @alicebob thanks for opening a PR - I think we need a tiny logic update to catch all the cases.

[slackpad]

I don't think this'll catch all cases since you won't always get an error. It's probably better to add something down on line 88 like this:

if p.lastIndex < oldIndex {
    p.lastIndex = 0
}

This'll reset the blocking query whenever the index slips backwards.

[alicebob]

Thanks for your reply, @slackpad.

The original fix I can easily reproduce in a test setup (watch() something and restart the Consul server in -dev mode), but a index which goes back (but doesn't drop the connection) I can't test. So I've added the lines you suggested, but I can't test it.

I also changed my original fix to reset the index to 0, not to 1.

(if this is fine I'll squash all commits)

[alicebob]

Thanks!