performance/

README

Knative Eventing Performance Tests

Configuring your cluster to run a benchmark

1. Create a namespace perf-eventing if it does not exist. To use a different namespace, please replace all the namespaces in all bash commands and yaml configuration files with your choice.

2. Install Knative eventing and components used in the performance test, such as MT broker, by following the steps in https://github.com/knative/eventing/blob/main/DEVELOPMENT.md.

3. Create a ConfigMap called config-mako in your chosen namespace containing the Mako config file.

kubectl create configmap -n perf-eventing config-mako --from-file=test/performance/benchmarks/<benchmark>/dev.config

1. Optionally edit the ConfigMap to set additional keys.

   kubectl edit configmap -n perf-eventing config-mako
   

NewConfigFromMap determines the valid keys in this ConfigMap. Current keys are:

• environment: Select a Mako config file in the ConfigMap. E.g. environment: dev corresponds to dev.config.
• additionalTags: Comma-separated list of tags to apply to the Mako run.

Run benchmarks continuously in CI using Mako

To run a benchmark continuously, and make the result available on Mako:

1. Use ko to apply yaml files in the benchmark directory.

ko apply -f test/performance/benchmarks/broker-imc/200-broker-imc-continuous-load-setup.yaml

Run without Mako

To run a benchmark once, and use the result from mako-stub for plotting:

1. Install the eventing resources for attacking:

   ko apply -f test/performance/benchmarks/broker-imc/100-broker-perf-setup.yaml
   

2. Start the benchmarking job:

   ko apply -f test/performance/benchmarks/broker-imc/300-broker-imc-increasing-load-setup.yaml
   

3. Wait until all the pods in namespace perf-eventing are completed.

4. Retrieve results from mako-stub using the script in knative/pkg where pod_name is the name of the aggregator pod:

   bash "$GOPATH/src/knative.dev/eventing/vendor/knative.dev/pkg/test/mako/stub-sidecar/read_results.sh" "$pod_name" perf-eventing ${mako_port:-10001} ${timeout:-120} ${retries:-100} ${retries_interval:-10} "$output_file"
   

   This will download a CSV with all raw results. Alternatively you can remove the port argument -p in mako-stub container to dump the output to container log directly.

Available benchmarks

• direct: Source -> Sink (baseline test)
• broker-imc: Source -> Broker with IMC -> Sink
• channel-imc: Source -> IMC -> Sink

Plotting results from mako-stub

In order to plot results from the mako-stub, you need to have installed gnuplot.

Three plot scripts are available:

• Only send/receive latencies
• Only send/receive throughput
• Combined send/receive throughput

To use them, you need to pass as first parameter the csv. If you want to use the combined plot script, you need to specify also latency upper bound, thpt lower and upper bound to show. For example:

gnuplot -c test/performance/latency-and-thpt-plot.plg data.csv 0.5 0 1100

• 0.5 is the time in seconds, and it is the max allowed size for the y1 axis
• 0 and 1100 are the message throughput, and it they represent the min and max boundaries of the y2 axis

Profiling

Most eventing binaries under cmd package are bootstrapped by either sharedmain.Main in knative.dev/pkg/injection/sharedmain or adapter.Main in knative.dev/eventing/pkg/adapter. These Main helper functions uses the profiling package to enable golang profiling by reading the profiling.enable flag in the config-observability configmap.

To enable profiling,

1. Add or modify profiling.enable: "true" in config/config-observability.yaml's data field and apply the change. Or use kubectl edit configmap -n knative-eventing config-observability.
2. Port forward into the pod which you want to profile, e.g., kubectl port-forward <imc-dispatcher-pod> 8008:8008
3. Point your browser to http://localhost:8008/debug/pprof/ and view pprof data.

After you are done, you can disable profiling by setting profiling.enable: "false".