Two clusters needed, a target cluster that will be tested and one that will be used to generate the load
TLS routes will be used as they're the ones that generate more load in the servers. In the target cluster:
git clone https://github.com/rporres/http-ci-tests.git
cd http-ci-tests/content/quickstarts/nginx
for i in $(seq 0 24); do
oc process -p IDENTIFIER=$i -f server-tls-edge.yaml | oc create -f -
oc process -p IDENTIFIER=$i -f server-tls-reencrypt.yaml | oc create -f -
done
and create a routes.txt file:
oc get routes -o json | jq -r '.items[] | .spec["host"]' > routes.txt
It will hold the information to upload result files to S3
ACCESS_KEY_ID=xxxx
SECRET_ACCESS_KEY=xxxx
oc create secret generic aws-s3 \
--from-literal=access_key_id=$ACCESS_KEY_ID
--from-literal=secret_access_key=$SECRET_ACCESS_KEY
They will be used by mb to determine how to send requests to the target cluster
git clone https://github.com/rporres/mb-k8s.git
cd mb-k8s
./utils/build-request-json.sh -f targets.txt -k 1 -c 50 -d 0 -t > request-file-1ka-50c.json
oc create configmap --from-file=request-file-1ka-50c.json request-file-1ka-50c
./utils/build-request-json.sh -f targets.txt -k 10 -c 50 -d 0 -t > request-file-10ka-50c.json
oc create configmap --from-file=request-file-10ka-50c.json request-file-10ka-50c
./utils/build-request-json.sh -f targets.txt -k 100 -c 50 -d 0 -t > request-file-100ka-50c.json
oc create configmap --from-file=request-file-100ka-50c.json request-file-100ka-50c
We don't use more than 50 clients per load server to avoid congestion that could affect in any way to the results, but the experience shows it is a conservative setting for servers like m5.xlarge
git clone https://github.com/rporres/mb-k8s.git
cd mb-k8s
helm install <test-name> helm/mb-k8s -f <values.yaml>
where values.yaml is one of the files inside mb-k8s-values-files
The result of the tests is uploaded into s3BucketName. They have to be downloaded and for those tests that run into multiple servers, sorted into one file, e.g. the results of a test distributed in four servers
for i in $(ls 100ka-200c-mb-k8s-*.xz); do
xz -dc $i
done | sort -t, -k1n,1n | xz > 100ka-200c.xz
In order to create data files for gnuplot from the result files, a utility script is provided create-gnuplot-values-file.py. It will create data files counting request rate, median latencies and error rates. Example gnuplot files can be found in gnuplot directory
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent