Gearpump is now able to [process 18 million messages/second (100 bytes per message) with a 8ms latency on a 4-node cluster](http://www.gearpump.io/releases/latest/index.html#gearpump-performance).

## Change log:

- #1247, add authentication for UI
- #1855: when service is unavailable, do not query models without waiting.
- #1833 update performance related doc
- #1831, Services exits on bind failure
- #1779: show critical path latency for application 
- #1733 dynamic dag fall back to last dag version in case of Master HA
- #1804, fix navbar on mobile
- #1713 regression on HBaseSink
- #1838 YARN failure trying to contact YarnAppMaster when launched in YARN