CP-47363: Fit the clustered kuttl release on the Kind node

[evan-cz]

Stacked on #1004.

The clustered kuttl suite never completes its install, so the whole k8s version matrix fails — and since docker-build is a rollup that needs: k8s-version-matrix-tests, that blocks merging anything.

Cause

The suite installs a second full-size release alongside the default cz-agent one already on the node. At chart defaults that is another ~1.3 CPU and ~2Gi of requests, and clustered mode fans the server out to three replicas. Pods stay Pending and helm --wait expires.

                    before overlay          after overlay
ksm            x1   100m / 256Mi            100m / 256Mi
server         x3   100m / 320Mi   →   x1    100m / 320Mi
aggregator     x3   200m / 128Mi   →   x1     20m / 128Mi
webhook        x3   100m / 128Mi   →   x1    100m / 128Mi
                    ~1.3 CPU, ~2Gi         ~320m, ~832Mi

Fix

CP-43292 already solved this for the webhookServer suites and added tests/kuttl/webhookserver-min-footprint.yaml for exactly this reason. Its own comment names our case:

Resource pressure. The default release already requests ~1.7 CPU; a second full-size stack pushed the Kind node past its allocatable CPU, so the new pods stayed Pending. … defaults.replicas: 1 → one replica per component (clustered mode would otherwise fan the server out to 3).

Both webhook suites use that overlay. alloy-clustered-test never adopted it. This applies it.

This failure predates the stack

The suite's steps silently no-opped until the commands: conversion (77b42c73, 2026-06-17), and the only matrix run recorded since then failed. This job needs: docker-build, which the image scan has been failing for weeks, so the matrix has been skipped throughout. Clearing the scan is what made it run again and surfaced this.

Also included

State dump on failure — pod phases, container states, events, Alloy logs. kuttl deletes the namespace the moment a step fails, so every previous failure produced a bare context deadline exceeded and nothing else; that is why this went undiagnosed for three months. The wait also goes 3m → 8m, since at three minutes a slow start and a broken install are indistinguishable.

🤖 Generated with Claude Code

[greptile-apps]

Greptile Summary

The PR reduces the resource footprint of the clustered KUTTL release and improves install-failure diagnostics.

• Applies the existing minimum-footprint Helm overlay to the clustered test release.
• Extends the install wait and command timeout for cold Kind nodes.
• Dumps pod, container, event, and Alloy log state when installation fails.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
tests/kuttl/alloy-clustered-test/steps/01-install-clustered-chart.yaml	Applies the shared low-resource test overlay and adds guarded failure diagnostics without changing production chart behavior.

_{Reviews (5): Last reviewed commit: "CP-47363: Make the clustered kuttl insta..." | Re-trigger Greptile}