Optionally _clearBuffer on _reportErrorStreak Threshold

[dm03514]

We have NGINX proxying to our satellites and nginx has a default max payload size of 1MB. We've seen:

• Satellites were not accepting traffic for a brief amount of time
• Client buffer filled up (Somewhere in here Satellites restarted)
• Satellites start accepting traffic again
• By this point client buffer was too big with the default span count limit of 4096
• Client tried to flush to satellites
• Rejected by NGINX config of max payload size of 1MB
• Buffer stays full as more spans are reported but can’t flush due to NGINX limit, hence increased memory usage

There looks to be a clean path to implementing this.

• https://github.com/lightstep/lightstep-tracer-javascript/blob/master/src/imp/tracer_imp.js#L1085
• https://github.com/lightstep/lightstep-tracer-javascript/blob/master/src/imp/tracer_imp.js#L814

---

If possible it would also be nice to have an option to set max buffer size bytes:

• Configure max buffer size bytes
• if a span would go over the size would flush I'm not sure of the performance implications of serializing across the different protocols. Ie are records concatenated so each span can be serialized and added to each other? or is there a different data structure that requires an array of spans making it not possible to do this??