ClusterBuster is a command line tool to run a variety of workloads on OpenShift 4 clusters. It was created with the following goals in mind:

• Stress the cluster: it was originally written in 2019 to simplify running scalable workloads as part of testing to 500 pods per node. It now supports many additional workloads, but all of them can be used for stress testing.

• Benchmarking clusters: this tool is used by the Red Hat Performance and Scalability organization to benchmark clusters. It collects performance data about each benchmark it runs, and also collects Prometheus metrics data.

• Command-line focus and scriptability: ClusterBuster is a traditional Linux command line tool. It is script-friendly and easy to use as a component of a larger test system. One of the tools included is a performance CI suite that runs ClusterBuster many times and packages the results. If you prefer using the command line to writing YAML, ClusterBuster may be the tool for you.

• Modularity: it is possible to add additional workloads by simply adding a few files.

• Reporting: ClusterBuster generates reports for both single runs and multiple ones, creating text output that can be further processed with traditional command line tools.

ClusterBuster supports all workloads running in pods, Red Hat OpenShift Virtualization VMs, and OpenShift sandboxed containers. ClusterBuster arranges for orchestration of workloads and synchronization between all pods/VMs so that all benchmark steps start simultaneously.

ClusterBuster currently supports the following workloads:

• cpusoaker -- a simple loop in Python that can be used either as a very rough measure of CPU performance or a load generator.

• files -- create, read, and delete large numbers of files to stress filesystem handling.

• fio -- the Flexible I/O tester for testing storage performance. This is not a fully general front end to fio; it supports basic I/O patterns.

• logger -- generate log messages.

• memory -- allocate, free, and optionally use large chunks of memory.

• server -- client-server message exchange.

• sysbench -- a multi-threaded system stress/benchmark tool.

• hammerdb -- HammerDB TPC-C and TPROC-C database benchmark (NOPM/TPM). Client and database run colocated (same pod or VM). Supports PostgreSQL and MariaDB. Not supported on arm64 (no package shipped yet; see HammerDB #767). Use the dedicated clusterbuster-hammerdb container and clusterbuster-hammerdb-vm VM images (see lib/container-image/).

• uperf -- a network performance tool front end.

In addition, these special purpose workloads are available:

• byo -- run a workload of your choice under ClusterBuster.

• failure -- run pods that deliberately fail. Intended for testing ClusterBuster itself.

• pausepod -- run a minimal pause pod that sleeps forever and cannot be used for other purposes.

• sleep -- sleep for configured interval

• synctest -- synchronization-only workload for validating sync infrastructure.

• waitforever -- dummy test, runs a workload that waits forever. This can be used to create pods or VMs that persist as infrastructure for running non-ClusterBuster workloads.

Workload plugins are under lib/clusterbuster/driver/workloads.

The original bash implementation has been renamed to clusterbuster.sh and is deprecated. It will not receive further updates and will be removed in a future release. Please use the new Python-based ./clusterbuster driver instead, which provides the same CLI interface with improved maintainability and test coverage.

The repository includes a pyproject.toml so you can install ClusterBuster in editable mode:

pip install -e .

Declared dependencies include PyYAML, kubernetes, and openshift (dynamic client). OpenShift-specific behavior (for example Prometheus under openshift-monitoring) is preserved.

The driver (./clusterbuster at the repo root) provides full CLI parsing, YAML and legacy job-file support, dry-run output, and orchestration of all supported workloads.

Job files may use either YAML format (an options: mapping with dash-separated keys and standard true/false values) or the legacy line-oriented format. YAML examples are in examples/.

Perf CI orchestration is Python via run-perf-ci-suite and lib/clusterbuster/ci/ (the legacy bash CI driver and lib/CI/workloads/*.ci fragments have been removed).

• lib/clusterbuster/driver/ — driver: config, CLI, orchestrator, sync, monitoring, reporting, metrics, cleanup, artifacts, manifests.
• lib/clusterbuster/driver/workloads/ — workload plugins (15 workloads, WorkloadBase + @register decorator).
• lib/clusterbuster/ci/ — CI compatibility layer (run-perf-ci-suite support).
• lib/pod_files/ — in-pod workload scripts (clusterbuster_pod_client subclasses).
• tests/ — pytest suite (529+ tests).

Please peruse the full documentation