Stress-API

This project aims to provide Kubernetes cluster operators with means to stress test newly built clusters with arbitrary workload. Project tries to be as simple as possible - it's basically just an API over Stress-ng with simple key authentication.

Basic usage

Add helm repo:

    $ helm repo add a-a-abramov https://a-a-abramov.github.io/stress-api
    $ helm repo update

Application needs a non empty apiSecret to work! Please set it during installation. Here we don't create an Ingress resource for our ClusterIP service. But if your cluster already has an Ingress Controller you're free to create one - values.yaml has an example.

    $ helm install --namespace <ns> --set apiSecret=<apikey> <name> a-a-abramov/stress-api

In case your cluster doens't have any ingress yet you may just kubectl proxy to access Stress-API:

    $ kubectl proxy &
    $ export ep=http://localhost:8001/api/v1/namespaces/<ns>/services/<name>-stress-api:80/proxy/v1.0/runjob
    $ curl -s -X 'POST' $ep -H 'Accept: application/json' -H 'X-Auth-Key: <apiKey>'  -H 'Content-Type: text/plain'  --data-binary $'cpu 1\ntimeout 1s' | jq

This request runs 1 CPU stressor for 1 second. Successful output:

    {
        "info": {
            "system-info": {
            "bufferram": 42872832,
            "cpus": 4,
            "cpus-online": 4,
            "date-yyyy-mm-dd": 7279455,
            "epoch-secs": 1650058434,
            "freeram": 6009737216,
            "freeswap": 1072676864,
            "hostname": "snga-stress-api-779866b64d-6rlzk",
            "machine": "x86_64",
            "nodename": "snga-stress-api-779866b64d-6rlzk",
            "pagesize": 4096,
            "release": "5.10.104-linuxkit",
            "run-by": "uwsgi",
            "sharedram": 351858688,
            "stress-ng-version": "0.13.05",
            "sysname": "Linux",
            "ticks-per-second": 100,
            "time-hh-mm-ss": 77634,
            "totalram": 8346984448,
            "totalswap": 1073737728,
            "uptime": 2417,
            "version": "#1 SMP Wed Mar 9 19:05:23 UTC 2022"
            }
        },
        "retcode": 0
    }

Stressors format

There are a couple of examples in the repo, but for any advanced use you have to familiarize yourself with stress-ng job file format from the tool's repository.

In order to run a job you should either define it inline:

    $ curl ... --data-binary $'cpu 1\ntimeout 1s'

Or specify absolute or relative path to job file:

    $ curl ... --data-binary '@jobs/cpu-0-15s.job'

Sample linear RAM workload generator

This example will start a new job that just maps 24M of memory and waits for 5 minutes (300 seconds) each 30 seconds for an hour. Simple dummy workloads like this can be used to test Horizontal Pod Autoscaler and/or Cluster Autoscaler.

    for i in `seq 1 120`
    do
        echo "Job $i"

        curl -X 'POST' <Cluster URI>/v1.0/runjob \
        -H "Accept: application/json" \
        -H "X-Auth-Key: <apikey>" \
        -H "Content-Type: text/plain" \
        --data-binary "@stress-api/jobs/ram-1-24M-300s.job" &

        sleep 30
    done