菜单
Run a storage load test against the kube-shard secondary API server by creating large PipelineRuns. Use when the user asks to load test, stress test, or benchmark the shard's storage capacity.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/konflux-ci/kube-shard --skill load-test命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | load-test |
| description | Run a storage load test against the kube-shard secondary API server by creating large PipelineRuns. Use when the user asks to load test, stress test, or benchmark the shard's storage capacity. |
Stress-test the kube-shard secondary API server by creating many large PipelineRuns that the Tekton controller processes into TaskRuns, multiplying the actual storage footprint.
kubectl configured to talk to the cluster# 1000 PipelineRuns, 100KB each, 10 tasks per PR, 10 in parallel
./hack/loadtest/run-loadtest.sh 1000 100 10 10
| Position | Env Var | Default | Description |
|---|---|---|---|
$1 | - | 1000 | Number of PipelineRuns to create |
$2 | - | 100 | Size of each PipelineRun in KB |
$3 | - | 10 | Number of tasks per PipelineRun |
$4 | - | 10 | Parallel kubectl processes |
| - | NAMESPACE | default | Target namespace |
| - | PIPELINERUN_YAML | (generated) | Path to pre-generated YAML |
Each PipelineRun with N tasks generates:
A 100KB PipelineRun with 10 tasks produces roughly 11x the raw PR size in total DB writes (including Kine's revision history).
| Goal | Command |
|---|---|
| Quick smoke test | ./hack/loadtest/run-loadtest.sh 100 100 10 10 |
| Exceed 8GB etcd limit | ./hack/loadtest/run-loadtest.sh 1000 100 10 10 |
| Heavy stress test | ./hack/loadtest/run-loadtest.sh 5000 100 10 10 |
While the load test runs, monitor PostgreSQL:
kubectl exec deployment/konflux-tekton-shard-postgresql -n kube-shard-operator -- \
psql -U kine -d kine -c "SELECT pg_size_pretty(pg_database_size('kine')) AS db_size, count(*) AS rows FROM kine;"
Watch for OOM kills on Kine and apiserver pods:
kubectl get pods -n kube-shard-operator -l app.kubernetes.io/instance=<shard-name> -w
If Kine pods OOM, bump memory via the APIShard CR:
kubectl patch apishard <name> --type=merge -p '{"spec":{"kine":{"resources":{"limits":{"memory":"16Gi"}}}}}'
kubectl delete pipelineruns --all -n default --wait=false
kubectl delete taskruns --all -n default --wait=false
After deletion, run a PostgreSQL VACUUM to reclaim disk:
kubectl exec deployment/konflux-tekton-shard-postgresql -n kube-shard-operator -- \
psql -U kine -d kine -c "VACUUM FULL kine;"
hack/loadtest/generate-pipelinerun.py -- generates a single PipelineRun
YAML of a target size with random non-compressible paddinghack/loadtest/run-loadtest.sh -- batch-creates PipelineRuns and reports
progress with estimated storage usage