load-test

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.

Storage Load Test

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.

Prerequisites

• kube-shard operator deployed with an APIShard CR active
• Tekton controller running and connected to the shard
• PostgreSQL pod accessible for DB size checks
• kubectl configured to talk to the cluster

Quick Start

# 1000 PipelineRuns, 100KB each, 10 tasks per PR, 10 in parallel
./hack/loadtest/run-loadtest.sh 1000 100 10 10

Parameters

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

Storage Multiplier

Each PipelineRun with N tasks generates:

• 1 PipelineRun object (written + updated multiple times)
• N TaskRun objects (each written + updated)
• Status updates on all objects

A 100KB PipelineRun with 10 tasks produces roughly 11x the raw PR size in total DB writes (including Kine's revision history).

Sizing Guide

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

Monitoring DB Size

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;"

Monitoring Pod Health

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"}}}}}'

Cleanup

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;"

Scripts

• hack/loadtest/generate-pipelinerun.py -- generates a single PipelineRun YAML of a target size with random non-compressible padding
• hack/loadtest/run-loadtest.sh -- batch-creates PipelineRuns and reports progress with estimated storage usage