Skip to content

Latest commit

 

History

History
117 lines (94 loc) · 5.54 KB

README.md

File metadata and controls

117 lines (94 loc) · 5.54 KB

Benchmark Operator

The intent of this Operator is to deploy common workloads to establish a performance baseline of Kubernetes cluster on your provider.

Workloads status

Workload Use Status in Operator Reconciliation usage VM support (kubevirt) Kata Containers
UPerf Network Performance Working Used, default : 3second Working Working
Iperf3 Network Performance Working Used, default : 3second Not Supported Preview
fio Storage IO Working Used, default : 3second Working Working
Sysbench System Performance Working Used, default : 3second Not Supported Preview
YCSB Database Performance Working Used, default : 3second Not Supported Preview
Byowl User defined workload Working Used, default : 3second Not Supported Preview
Pgbench Postgres Performance Working Used, default : 3second Not Supported Preview
Smallfile Storage IO Performance Working Used, default : 3second Not Supported Preview
fs-drift Storage IO Longevity Working Not used Not Supported Preview
hammerdb Database Performance Working Used, default : 3second Not Supported Preview
Service Mesh Microservices Working Used, default : 3second Not Supported Preview
Vegeta HTTP Performance Working Used, default : 3second Not Supported Preview
Scale Openshift Scale Openshift Cluster Working Used, default : 3second Not Supported Preview
stressng Stress system resources Working Used, default: 3second Not Supported Preview
kube-burner k8s Performance Working Used, default : 3second Not Supported Preview
cyclictest Real-Time Performance Working Used, default : 3second Not Supported Preview
oslat Real-Time Latency Working Used, default : 3second Not Supported Preview

Reconciliation

Previously the Benchmark Operator didn't properly take advantage of the reconciliation period. Going forward we will make every attempt to utilize the reconciliation period.

Why did we decide to switch to this? Our operator would implement long running tasks, due to the nature of benchmarks. However, long running tasks blocks the Operator, causing us to delete the Operator and re-create the operator to un-block it. The benchmarks mentioned above that state Used for Reconciliation, no longer have this issue.

Optional workload images

Optional locations for workload images can now be added easily without the need to rebuild the operator. To do so in the workload args section of the CR add image: [location]

NOTE: This is not a required arguement. If omitted it will default to the currently verified workload image. Additionally, this is NOT enabled for YCSB

For Example:

apiVersion: ripsaw.cloudbulldozer.io/v1alpha1
kind: Benchmark
metadata:
  name: example-benchmark
  namespace: my-ripsaw
spec:
  elasticsearch:
    server: "my-es.foo.bar"
    port: 9200
  metadata_collection: true
  cleanup: false
  workload:
    name: "foo"
    args:
      image: my.location/foo:latest

User Provided UUID

All benchmarks in the benchmark-operator utilize a UUID for tracking and indexing purposes. This UUID is, by default, generated when the workload is first started. However, if desired, a user provided UUID can be added to the workload cr.

NOTE: The provided UUID must be in format XXXXX-XXXXX-XXXXX

For Example:

apiVersion: ripsaw.cloudbulldozer.io/v1alpha1
kind: Benchmark
metadata:
  name: example-benchmark
  namespace: my-ripsaw
spec:
  uuid: 6060004a-7515-424e-93bb-c49844600dde
  elasticsearch:
    server: "my-es.foo.bar"
    port: 9200
  metadata_collection: true
  cleanup: false
  workload:
    name: "foo"
    args:
      image: my.location/foo:latest

Installation

Installation

Contributing

Contributing

Metadata Collection

Metadata Collection

Cerberus Integration

Cerberus Integration

Indexing to Elasticsearch

Indexing to Elasticsearch

Capturing Prometheus Data

Capturing Prometheus Data

Cache dropping

Cache dropping

Community

Key Members(slack_usernames): ravi, mohit, dry923, rsevilla or rook