Hi folks, this is somewhat of a follow-up to my issue here but reproduced in minikube so, hopefully, it’s a little easier to discuss exact setup details. You can find my project here with reproduction steps in the README. My problem is as follows. I have three services
-
obfuscator
: a gRPC service that reverses whatever you send it -
helloworld
: an HTTP service that sends ‘hello, world’ toobfuscator
and then wraps the result up in json -
benchmarker
: a service to generate load againsthelloworld
with the first two emitting opencensus traces directly to a central jaeger collector, the final relying entirely on the mesh traces for instrumentation. What I find after letting the cluster run for a while is that the appears to be a distinction between service names if the traces are istio sourced versus opencensus sourced. Unfortunately I am a new user and cannot post many images. You can find screenshots in my reproduction project under img/
.
Opencensus originated traces are connected:
but istio originated traces do not appear to be.
Worth pointing out that the last two screenshots do show benchmarker.spangen
collected in the trace. If you look at my jaeger configuration under k8s/manifests/jaeger/jaeger.yaml
you’ll notice that it’s an in-memory setup but this issue is something I’ve managed to reproduce from my production cluster with disk-backed storage of traces. I have got istio pointed to jaeger like so and am working with istio 1.3.1, per the install script just linked.
Does anyone have a notion of what might be the matter?