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:
- hammerdb: running hammerdb workload on the following databases: MSSQL, Mariadb, Postgresql in Pod, Kata or VM with Configuration
- stressng: running stressng workload in Pod, Kata or VM Configuration
- uperf: running uperf workload in Pod, Kata or VM with Configuration
- vdbench: running vdbench workload in Pod, Kata or VM with Configuration
- bootstorm: calculate VMs boot load time Configuration
** 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 is available at benchmark-runner.readthedocs.io
Table of Contents
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 $KUBECONFIG:/root/.kube/config --privileged quay.io/ebattat/benchmark-runner:latestor
podman run --rm -e WORKLOAD=$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:latestSAVE RUN ARTIFACTS LOCAL:
- add
-e SAVE_ARTIFACTS_LOCAL='True'or--save-artifacts-local=true - add
-v /tmp:/tmp - git clone -b v1.0.2 https://github.com/cloud-bulldozer/benchmark-operator /tmp/benchmark-operator
[TBD]
There are 2 grafana dashboards templates:
** After importing json in grafana, you need to configure elasticsearch data source. (for more details: see HOW_TO.md)
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.
see HOW_TO.md