Negative hex spanIds

[henrikno]

Currently you can get negative span ids like this: -140dcf65fae75f53
I don't think this is compatible with zipkin, e.g. if a zipkin/finagle client calls a brave service (tell me if I'm wrong here).

I think we should use Long.toHexString when we serialize span/traceIds, and Long.parseUnsignedLong when deserializing them, or some similar method.
See how finagle generates ids:
https://github.com/twitter/finagle/blob/master/finagle-core/src/main/scala/com/twitter/finagle/tracing/Id.scala

[joeslice]

I believe this was fixed by @srapp with #28, released in 2.2.1. Are you using a version prior to that?

[kristofa]

Hi @henrikno. There indeed was a fix as @joeslice mentions but I I guess it would be good if we would write a test that verifies and confirms if we still have an incompatibility between Finagle and Brave? Could you do that? Thanks!

[kristofa]

@henrikno you are right. I wrote a test and we are incompatible with zipkin / finagle when negative span ids are generated. I'll look into you suggestions.

[kristofa]

I can fix it by using following guava parsing/conversion utilities (part of com.google.common.primitives package):

From Long to String: UnsignedLongs.toString(spanId, 16)
From String to Long: UnsignedLongs.parseUnsignedLong(stringSpanId, 16)

java.lang.Long.toHexString(spanId) for converting to String also works but I prefer to stick to Guava for both conversions.

This makes it compatible with Zipkin/Finagle. I tested with positive and negative long values.
I'll fix this and close this issue when it is done.

Apparently this bug is fixed in Java 8.

[kristofa]

This issue is fixed as part of brave 2.4 which is just released and should become available through Maven Central in a few hours.