Skip to main content

​
Overview

Dev-fast mode (bootstrap.sh) is optimized for daily development with instant warm starts using hash-based caching. It uses kindnetd CNI (no Cilium) for maximum speed and supports a single control-plane node.
Cold start: ~120s | Warm start: Instant (hash match) / 10-15s (manifest changes)

​
Key Features

  • kindnetd CNI - No Cilium overhead for faster startup
  • Single node - Control-plane only, minimal resource usage
  • Warm cluster support - Hash-based state detection
  • 4-phase parallel execution - Maximum concurrency
  • No Istio/ArgoCD - Lightweight for app development

​
Command Usage

# Standard bootstrap (warm cluster aware)
bootstrap

# Force clean rebuild
bootstrap --clean

# Delegate to Cilium mode
bootstrap --full

​
Flags

FlagDescription
--cleanDelete existing cluster and force cold start
--fullDelegate to bootstrap-full.sh (Cilium mode)

​
Warm Cluster (Hash Gate)

The bootstrap script uses SHA-256 hashes to detect cluster state and avoid unnecessary rebuilds:

​
Hash Storage

  • .bootstrap-state/cluster - Hash of kind-config-dev.yaml + images.sh
  • .bootstrap-state/manifest - Hash of entire manifests-result/ directory

​
Decision Logic

1

Cluster running + cluster hash match + manifest hash match

Instant complete - Health check only, no deployment
2

Cluster running + cluster hash match + manifest hash mismatch

Warm reapply (~10-15s) - Regenerate and reapply manifests only
3

Cluster hash mismatch or cluster stopped

Cold start (~120s) - Full cluster rebuild

​
4-Phase Parallel Execution

​
Phase 1: Preparation (Parallel)

All tasks run concurrently:
  • kind-cluster - Create kind cluster with kind-config-dev.yaml
  • gen-manifests - Generate Kubernetes manifests
  • otel-fetch - Fetch OTel collector image (smart mode: R2 cache → local cache → build)
  • image-preload - Pull container images in parallel
# From bootstrap.sh:218-223
timed_step "phase1-prep" parallel_run \
  "kind-cluster:_step_kind_cluster" \
  "gen-manifests:bash ${SCRIPT_DIR}/gen-manifests.sh" \
  "otel-fetch:bash ${SCRIPT_DIR}/load-otel-collector-image.sh smart" \
  "image-preload:_step_image_preload"
Typical duration: ~12s

​
Phase 2: Image Load (Sequential)

Load images into kind cluster:
  • Load all preloaded images into kind
  • Load OTel collector image
Typical duration: ~37s

​
Phase 2.5: PostgreSQL Early Start

PostgreSQL starts early to overlap its long startup time (~87s) with other deployments. 
_step_postgresql_apply

​
Phase 3: Deploy Services (Parallel)

All services deploy concurrently:
  • garage - S3-compatible storage (wait for ready + setup)
  • observability - Prometheus stack, Loki, Tempo, OTel collector
  • traefik - Ingress controller + auth
  • cloudflared - Cloudflare tunnel (if credentials exist)
# From bootstrap.sh:232-236
timed_step "phase3-deploy" parallel_run \
  "garage:_step_garage_deploy" \
  "observability:_step_observability" \
  "traefik:_step_traefik" \
  "cloudflared:_step_cloudflared"
Typical duration: ~21s

​
Phase 4: Wait for Pods (Parallel)

Wait for critical pods in parallel:
  • PostgreSQL (database namespace)
  • Grafana (observability namespace)
  • Prometheus (observability namespace)
Typical duration: ~52s

​
Benchmark Results

Total Time

~122s cold start

vs Cilium Mode

38% faster (~197s → ~122s)
From Apple M4, 24GB RAM:
PhaseTime
phase1-prep~12s
phase2-load~37s
phase3-deploy~21s
phase4-wait~52s
TOTAL~122s

​
Cluster Configuration

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
name: microservice-infra
nodes:
  - role: control-plane
    kubeadmConfigPatches:
      - |
        kind: InitConfiguration
        nodeRegistration:
          kubeletExtraArgs:
            node-labels: "ingress-ready=true"
    extraPortMappings:
      - containerPort: 30300  # Grafana
        hostPort: 30300
      - containerPort: 30081  # Traefik
        hostPort: 30081
      - containerPort: 30090  # Prometheus
        hostPort: 30090
      - containerPort: 30093  # Alertmanager
        hostPort: 30093
No Cilium: Uses default kindnetd CNI for simplicity and speed.

​
Exposed Services

After bootstrap completes:
ServiceURLCredentials
Grafanahttp://localhost:30300admin/admin
Prometheushttp://localhost:30090-
Alertmanagerhttp://localhost:30093-
Traefikhttp://localhost:30081-

​
Next Steps

After dev-fast bootstrap: 
cd microservice-app && tilt up

​
Use Cases

Daily Development

Instant restarts with warm cluster support

Quick Testing

Fast iteration cycles without Cilium overhead

Resource Constrained

Single node, minimal footprint

App Development

Full observability stack without service mesh

​
Comparison

See the Bootstrap Mode Comparison to understand when to use dev-fast vs other modes.

Build docs developers (and LLMs) love

Get started for freeTalk to us