It is challenging to provide HA for syslog because the syslog protocol itself was not designed with HA as a goal. See Performant AND Reliable Syslog UDP is best for an excellent overview of this topic.
The gist is that the protocol itself limits the extent to which you can make any syslog collection architecture HA; at best it can be made “mostly available”. Think of syslog as MP3 – it is a “lossy” protocol and there is nothing you can do to restore it to CD quality (lossless). Some have attempted to implement HA via front-side load balancers; please don’t! This is the most common architectural mistake folks make when architecting large-scale syslog data collection. So – how to make it “mostly available”? Keep it simple, and use OS clustering (shared IP) or even just VMs with vMotion. This simple architecture will encounter far less data loss over time than more complicated schemes.
