Repository for dynamic tracing queries.

Dependencies

This repository relies on Minikube and GCP to emulate microservice scenarios. We primarily use Ubuntu 18.04 for our tests. For convenience, dependencies can be installed with the ./tools/setup.sh command. It contains all the steps necessary to set up the compiler and testing environment.

Package Dependencies (Minikube)

• rust to build the dynamic tracing compiler
• kubernetes to administrate and run web applications in a cluster
• minikube to run a local Kubernetes cluster
• docker as the container driver back end of Minikube
• bazel to build Wasm filters for Envoy
• istio to manage Envoy and its filters
• python3.6 to run the scripts

Additional Package Dependencies (GCP)

TBD

Python Dependencies

• prometheus-api-client to query Prometheus

The Dyntracing Compiler (dtc)

Building the Compiler

The compiler can be build with cargo build.

Compiling a Filter from a Query

After the compiler has been built, you can compile queries with the target/debug/dtc -q [QUERY] -u [USER_DEFINED_FUNCTION] command in the tracing compiler directory. For example, debug/dtc -q example_queries/count.cql -u example_udfs/count.cc compiles a counting query with the user defined function "count". The result will be stored in cpp_filter/filter.cc.

The Testing Environment

Environment Setup

Once everything is installed, the Kubernetes cluster can be started with ./kube_env.py --setup if you are running on minikube or ./kube_env.py --setup -p GCP if you are running on google cloud. If you are running on google cloud, add -p GCP to all subsequent commands as well. Starting the cluster will take a few minutes.

Building a filter

We provide a pre-built filter. It can be deployed with the ./kube_env.py --deploy-filter command. If the filter was generated from the compiler, it can be rebuilt with the ./kube_env.py --build-filter command

Testing the filter

Once the filter has been successfully installed, it is possible to run experiments with ./run_experiments.py --num-experiments 1. You can also issue single HTTP requests with ./env/send_request.py.

Teardown

Remove the filter

./kube_env.py --undeploy-filter

Remove the deployment

./kube_env.py --remove-bookinfo

Tear down the cluster

./kube_env.py --clean