Skip to content

CLAUDE.md

Latest commit

 

History

History
70 lines (53 loc) · 1.99 KB

CLAUDE.md

File metadata and controls

70 lines (53 loc) · 1.99 KB

instant-env

Bash spike to measure EC2 deployment speed. Learning rig before applying to brownfield backend.

Quick Reference

# One-time setup
./scripts/setup-security-group.sh
./scripts/setup-keypair.sh

# Run cold baseline benchmark
./bench/cold.sh ~/.ssh/instant-env-admin.pem

# Cleanup everything
./scripts/cleanup-all.sh

Directory Structure

bench/              # Benchmark scripts
  common.sh         # Shared functions (timing, SSH probe)
  cold.sh           # Baseline: full cloud-init
  start-stopped.sh  # Pre-warmed EBS/ENI (simulates warm pool)
  minimal-ami.sh    # Stripped AMI, no cloud-init
  custom-init.sh    # TODO: Go binary init
scripts/            # AWS setup (idempotent)
docs/               # Documentation

What We're Measuring

T0: API request
T1: RunInstances returns (instance ID)
T2: Instance state = "running"
T3: TCP port 22 open
T4: SSH auth succeeds

Target: Get T4-T0 under 10 seconds (baseline is 30-60s).

Techniques

Technique Description Result
cold Full RunInstances + cloud-init 25.8s (baseline)
start-stopped Pre-warmed EBS/ENI 21.2s (-18%)
hibernate Resume from hibernation 20.7s (-20%)
minimal-ami Stripped AL2023, no cloud-init 28.8s (+12% slower!)
warm-pool Running instance + SSH key inject 1.9s (-92%)
warm-pool-eic Running instance + EC2 Instance Connect 3.9s (-85%)
warm-pool-realistic EIC + EC2 tag-based pool mgmt 12.8s (-50%)

Sweep Results

Factor Variance Finding
Instance type 680ms t3→r7i all ~22-23s; Pending→Running fixed at 16.4s
Region 2116ms us-east-2→eu-west-1; Pending→Running fixed globally

Conclusion: Instance type and region don't matter. Warm pools are the only lever.

Conventions

  • All scripts use set -euo pipefail
  • Timing in milliseconds
  • Instance type fixed to m7i.large
  • All resources tagged Project=instant-env