Jaeger Tracing Looks Problematic

[Orfeasfil]

Hi there!

I deployed DeathStarBench on a Kubernetes cluster using helm. We initialized the social network graph and used the wrk command to run the following services:

1. read-home-timeline
2. compose-post-service
3. read-user-timeline

After running these services, we attempted to retrieve the traces via the Jaeger UI. The results seem problematic. For each service, a random request yields the following results:

read-home-timeline: command: /wrk2/wrk -D exp -t 100 -c 1000 -d 5 -L -s ./wrk2/scripts/social-network/read-home-timeline.lua http://${NGINX-THRIFT-SERVICE-IP}:8080/wrk2-api/home-timeline/read -R 100 results:

• Total latency of the request: 394.71ms
• Total latency of the services: 3.12ms

The latencies shown do not add up correctly, which seems erroneous.

compose-post-service: command: /wrk2/wrk -D exp -t 10 -c 1000 -d 5 -L -s ./wrk2/scripts/social-network/compose-post.lua http://${NGINX-THRIFT-SERVICE-IP}:8080/wrk2-api/post/compose -R 100 results:

read-user-timeline: command: /wrk2/wrk -D exp -t 100 -c 1000 -d 5 -L -s ./wrk2/scripts/social-network/read-user-timeline.lua http://${NGINX-THRIFT-SERVICE-IP}:8080/wrk2-api/user-timeline/read -R 100 results:

Troubleshooting Attempts: We experimented with different sampleParam values for Jaeger (0.01, 0.1, and 1), but the results remained consistent with the same discrepancies.

Steps to Reproduce:

1. Create a Kubernetes cluster.
2. Follow the instructions to initialize the social graph.
3. Run the read-home-timeline, compose-post-service, and read-user-timeline services.
4. Retrieve the traces via the Jaeger UI.

Do you have any idea on whats happening? Any suggestions or insights would be greatly appreciated.

[ccs1110]

it looks like that you delployed the microservices in different devices，and the transfer latency is bigger than service latency