[warp-agent-staging]

Automatically prefer native Kubernetes image volumes when backend.kubernetes.use_image_volumes is omitted, while honoring explicit opt-in and opt-out.

Linear: REMOTE-3249

Behavior

• Preserve omitted, explicit true, and explicit false through YAML, runtime configuration, and Helm. The chart leaves the key unset by default.
• In automatic mode, reuse the real namespaced startup Job to exercise image-volume support. Any failed or inconclusive probe selects the safe default: no image volumes, with the existing legacy preflight required before startup succeeds.
• Explicit true still requires a successful image-volume preflight; explicit false skips that probe and uses the legacy copy path.
• Select the loading path per sidecar: read-only sidecars use image volumes when enabled; writable sidecars use root init containers and writable emptyDir mounts. Mixed tasks work in both automatic and explicit-on mode.

Detection limits

API-server version, discovery, and dry-run do not prove admission and node-runtime support. The existing 15-second runtime probe remains the capability check, without extra cluster-scoped permissions. A successful probe proves only its selected node/runtime, not every node in a heterogeneous pool. Writable sidecars still require permission to run root init containers.

Validation

• Mixed-sidecar regression failed before the fix and passed afterward through ExecuteTask; automatic fallback on an inconclusive API timeout is also covered.
• go test . ./internal/config ./internal/worker -count=1 — passed.
• go vet . ./internal/config ./internal/worker and go build ./... — passed.
• golangci-lint run . ./internal/config ./internal/worker --concurrency 2 --timeout=5m — passed (v2.12.2).
• Helm v4.1.3 lint and omitted/true/false rendering — passed on the unchanged chart.
• Final gofmt and git diff --check — passed.
• No live Kubernetes cluster was available; real mount/runtime behavior was not verified.

[warp-agent-staging]

This PR was generated with Warp.

Comment @warp-staging-factory on this PR to send it follow-up work.

  

[warp-agent-staging]

Overview

The rebase preserves the PR's tri-state configuration and startup-Job probe while retaining main's metadata checks and upgrade guidance. The remaining detection policy needs a human decision.

Concerns

• Important — ambiguous: internal/worker/kubernetes.go:901-910 selects legacy mode for the worker's lifetime after every probe error, including a 15-second scheduling timeout or registry/network failure. Supported clusters can therefore lose image volumes, or fail startup when the fallback requires disallowed root init containers; this is not evidence of unsupported runtime capability. I recommend falling back only for clear incompatibility and failing/retrying startup for inconclusive errors: should that replace the current best-effort fallback policy?

Verdict

Checks: build pass, affected-package tests pass, affected-package vet pass, Helm pass, CI pending, visual proof n/a.

Found: 0 critical, 1 important, 0 suggestions, 0 nits. Human decision needed; no approval or rejection.

Responding as eng-factory-slim: Open session · View in factory