Skip to content

kelchen123/benchmark_testing

Repository files navigation



Benchmark-Runner: Running benchmarks

Actions Status PyPI Latest Release Container Repository on Quay Coverage Status Documentation Status python License

What is it?

benchmark-runner is a containerized Python lightweight and flexible framework for running benchmark workloads on Kubernetes/OpenShift runtype kinds Pod, kata and VM.

This framework support the following embedded workloads:

** For hammerdb mssql must run once permission

Benchmark-runner grafana dashboard example:

Reference:

  • The benchmark-runner package is located in PyPi
  • The benchmark-runner container image is located in Quay.io

Documentation

Documentation is available at benchmark-runner.readthedocs.io

Table of Contents

Run workload using Podman or Docker

The following options may be passed via command line flags or set in the environment:

mandatory: WORKLOAD=$WORKLOAD

mandatory: KUBEADMIN_PASSWORD=$KUBEADMIN_PASSWORD

mandatory: $KUBECONFIG [ kubeconfig file path]

Choose one from the following list:

['stressng_pod', 'stressng_vm', 'stressng_kata', 'uperf_pod', 'uperf_vm', 'uperf_kata', 'hammerdb_pod_mariadb', 'hammerdb_vm_mariadb', 'hammerdb_kata_mariadb', 'hammerdb_pod_mariadb_lso', 'hammerdb_vm_mariadb_lso', 'hammerdb_kata_mariadb_lso', 'hammerdb_pod_postgres', 'hammerdb_vm_postgres', 'hammerdb_kata_postgres', 'hammerdb_pod_postgres_lso', 'hammerdb_vm_postgres_lso', 'hammerdb_kata_postgres_lso', 'hammerdb_pod_mssql', 'hammerdb_vm_mssql', 'hammerdb_kata_mssql', 'hammerdb_pod_mssql_lso', 'hammerdb_vm_mssql_lso', 'hammerdb_kata_mssql_lso', 'vdbench_pod', 'vdbench_kata', 'vdbench_vm', 'clusterbuster', 'bootstorm_vm']

** clusterbuster workloads: cpusoaker, files, fio, uperf. for more details see

auto: NAMESPACE=benchmark-operator [ The default namespace is benchmark-operator ]

auto: ODF_PVC=True [ True=ODF PVC storage, False=Ephemeral storage, default True ]

auto: EXTRACT_PROMETHEUS_SNAPSHOT=True [ True=extract Prometheus snapshot into artifacts, false=don't, default True ]

auto: SYSTEM_METRICS=False [ True=collect metric, False=not collect metrics, default False ]

auto: RUNNER_PATH=/tmp [ The default work space is /tmp ]

optional: PIN_NODE_BENCHMARK_OPERATOR=$PIN_NODE_BENCHMARK_OPERATOR [node selector for benchmark operator pod]

optional: PIN_NODE1=$PIN_NODE1 [node1 selector for running the workload]

optional: PIN_NODE2=$PIN_NODE2 [node2 selector for running the workload, i.e. uperf server and client, hammerdb database and workload]

optional: ELASTICSEARCH=$ELASTICSEARCH [ elasticsearch service name]

optional: ELASTICSEARCH_PORT=$ELASTICSEARCH_PORT

optional: CLUSTER=$CLUSTER [ set CLUSTER='kubernetes' to run workload on a kubernetes cluster, default 'openshift' ]

optional:scale SCALE=$SCALE [For Vdbench/Bootstorm: Scale in each node]

optional:scale SCALE_NODES=$SCALE_NODES [For Vdbench/Bootstorm: Scale's node]

optional:scale REDIS=$REDIS [For Vdbench only: redis for scale synchronization]

optional: LSO_DISK_ID=$LSO_DISK_ID [LSO_DISK_ID='scsi-<replace_this_with_your_actual_disk_id>' For using LSO Operator in hammerdb]

optional: WORKER_DISK_IDS=$WORKER_DISK_IDS [WORKER_DISK_IDS For ODF/LSO workloads hammerdb/vdbench]

For example:

podman run --rm --workload=$WORKLOAD --kubeadmin-password=$KUBEADMIN_PASSWORD --pin-node-benchmark-operator=$PIN_NODE_BENCHMARK_OPERATOR --pin-node1=$PIN_NODE1 --pin-node2=$PIN_NODE2 --elasticsearch=$ELASTICSEARCH --elasticsearch-port=$ELASTICSEARCH_PORT -v $KUBECONFIG:/root/.kube/config --privileged quay.io/ebattat/benchmark-runner:latest

or

podman run --rm -e WORKLOAD=$WORKLOAD -e KUBEADMIN_PASSWORD=$KUBEADMIN_PASSWORD -e PIN_NODE_BENCHMARK_OPERATOR=$PIN_NODE_BENCHMARK_OPERATOR -e PIN_NODE1=$PIN_NODE1 -e PIN_NODE2=$PIN_NODE2 -e ELASTICSEARCH=$ELASTICSEARCH -e ELASTICSEARCH_PORT=$ELASTICSEARCH_PORT -e log_level=INFO -v $KUBECONFIG:/root/.kube/config --privileged quay.io/ebattat/benchmark-runner:latest

SAVE RUN ARTIFACTS LOCAL:

  1. add -e SAVE_ARTIFACTS_LOCAL='True' or --save-artifacts-local=true
  2. add -v /tmp:/tmp
  3. git clone -b v1.0.2 https://github.com/cloud-bulldozer/benchmark-operator /tmp/benchmark-operator

Run vdbench workload in Pod using OpenShift

Run vdbench workload in Pod using Kubernetes

Run workload in Pod using Kubernetes or OpenShift

[TBD]

Grafana dashboards

There are 2 grafana dashboards templates:

  1. grafana/func/benchmark-runner-ci-status-report.json
  2. grafana/func/benchmark-runner-report.json

** After importing json in grafana, you need to configure elasticsearch data source. (for more details: see HOW_TO.md)

Inspect Prometheus Metrics

The CI jobs store snapshots of the Prometheus database for each run as part of the artifacts. Within the artifact directory is a Prometheus snapshot directory named:

promdb-YYYY_MM_DDTHH_mm_ss+0000_YYYY_MM_DDTHH_mm_ss+0000.tar

The timestamps are for the start and end of the metrics capture; they are stored in UTC time (+0000). It is possible to run containerized Prometheus on it to inspect the metrics. Note that Prometheus requires write access to its database, so it will actually write to the snapshot. So for example if you have downloaded artifacts for a run named hammerdb-vm-mariadb-2022-01-04-08-21-23 and the Prometheus snapshot within is named promdb_2022_01_04T08_21_52+0000_2022_01_04T08_45_47+0000, you could run as follows:

$ local_prometheus_snapshot=/hammerdb-vm-mariadb-2022-01-04-08-21-23/promdb_2022_01_04T08_21_52+0000_2022_01_04T08_45_47+0000
$ chmod -R g-s,a+rw "$local_prometheus_snapshot"
$ sudo podman run --rm -p 9090:9090 -uroot -v "$local_prometheus_snapshot:/prometheus" --privileged prom/prometheus --config.file=/etc/prometheus/prometheus.yml --storage.tsdb.path=/prometheus --storage.tsdb.retention.time=100000d --storage.tsdb.retention.size=1000PB

and point your browser at port 9090 on your local system, you can run queries against it, e.g.

sum(irate(node_cpu_seconds_total[2m])) by (mode,instance) > 0

It is important to use the --storage.tsdb.retention.time option to Prometheus, as otherwise Prometheus may discard the data in the snapshot. And note that you must set the time bounds on the Prometheus query to fit the start and end times as recorded in the name of the promdb snapshot.

How to develop in benchmark-runner

see HOW_TO.md