Skip to content

LIVE_DEMO.md

Latest commit

 

History

History
189 lines (130 loc) · 6.44 KB

LIVE_DEMO.md

File metadata and controls

189 lines (130 loc) · 6.44 KB

Live Demo Script

Automated feature demonstration for GCO (Global Capacity Orchestrator on AWS). Run live_demo.sh in a terminal during presentations to showcase the platform's capabilities in real time.

Table of Contents

What It Does

live_demo.sh is a single script designed to be run in a visible terminal during a live presentation. It walks through GCO's core capabilities automatically, with clear narration, pauses, and visually formatted output so the audience can follow along.

The script reads cdk.json to detect which optional features are enabled (schedulers, FSx, Valkey) and adapts its flow accordingly — it only demos what's actually deployed.

Prerequisites

  • GCO CLI installed (gco --version)
  • Infrastructure deployed (gco stacks deploy-all -y)
  • EKS endpoint set to PUBLIC_AND_PRIVATE (for kubectl access during demo)
  • Cluster access configured (./scripts/setup-cluster-access.sh)
  • kubectl working (kubectl get nodes)
  • jq installed (brew install jq / apt install jq)
  • Terminal with at least 120 columns and color support

Running the Demo

# From the repo root:
bash demo/live_demo.sh

# Or with a specific region override:
GCO_DEMO_REGION=us-west-2 bash demo/live_demo.sh

The script pauses between sections. Press Enter to advance.

Recording the Demo

record_demo.sh captures a non-interactive run of live_demo.sh as an animated GIF for embedding in READMEs and documentation.

Recording Prerequisites

Everything needed for live_demo.sh (see Prerequisites above), plus:

  • asciinema — records the terminal session as a .cast file

    brew install asciinema     # macOS
pip install asciinema      # pip
apt install asciinema      # Debian/Ubuntu

  • agg — converts .cast to animated GIF (optional — skipped gracefully if missing)

    brew install agg           # macOS
cargo install agg          # Rust/cargo

Recording

# Record and generate GIF (from repo root)
bash demo/record_demo.sh

# Record only the .cast file (skip GIF conversion)
SKIP_GIF=1 bash demo/record_demo.sh

# Custom dimensions and speed
DEMO_COLS=140 DEMO_ROWS=40 DEMO_SPEED=3 bash demo/record_demo.sh

The script runs its own preflight validation (tools, cluster access, disk space) before recording. Output files:

File Description
demo/live_demo.cast Asciinema recording (JSON text, replayable with asciinema play)
demo/live_demo.gif Animated GIF for embedding in READMEs

Re-recording After Changes

After editing live_demo.sh or lib_demo.sh, re-record:

bash demo/record_demo.sh

The GIF is embedded in both demo/README.md and the main README.md. Commit the updated .gif and .cast files.

Replay Without Re-recording

asciinema play demo/live_demo.cast

What the Script Covers

Section Feature Condition
1 Cost visibility — summary, regional breakdown, daily trend Always
2 Capacity discovery — GPU availability, region recommendation, auto-region SQS Always
3 Volcano scheduler — gang scheduling example volcano.enabled = true in cdk.json
4 Kueue scheduler — quota-based job queueing kueue.enabled = true in cdk.json
5 YuniKorn scheduler — app-aware fair scheduling yunikorn.enabled = true in cdk.json
6 Slurm operator — HPC batch scheduling slurm.enabled = true in cdk.json
7 FSx for Lustre — high-performance scratch storage fsx_lustre.enabled = true in cdk.json
8 Valkey cache — serverless K/V caching valkey.enabled = true in cdk.json
9 Inference endpoint — deploy, invoke, and teardown Always (skip with SKIP_INFERENCE=1)
10 EFS shared storage — persistent job outputs Always

Customization

  • Skip sections: Set environment variables to skip specific parts:

    SKIP_COSTS=1 bash demo/live_demo.sh         # Skip cost section
SKIP_CAPACITY=1 bash demo/live_demo.sh      # Skip capacity section
SKIP_SCHEDULERS=1 bash demo/live_demo.sh    # Skip all scheduler demos

  • Region: Override the auto-detected region:

    GCO_DEMO_REGION=eu-west-1 bash demo/live_demo.sh

  • Speed: Adjust the typing delay and pause duration:

    GCO_DEMO_FAST=1 bash demo/live_demo.sh      # Shorter pauses

Maintenance Guide

This script depends on the example manifests in examples/ and the gco CLI. When updating GCO, check the following:

  1. New schedulers or features added to cdk.json — Add a detection block and demo section in live_demo.sh. Follow the pattern of existing scheduler sections.
  2. Example manifest names changed — Update the corresponding submit commands in the script. The manifest filenames are referenced directly.
  3. CLI command changes — If gco subcommands change syntax, update the commands in the script. The script calls gco costs, gco jobs, and gco files directly.
  4. New example jobs — If a new example is added that showcases a feature worth demoing, add a section. Use the section_header, narrate, and run_cmd helper functions for consistent formatting.
  5. cdk.json schema changes — The script parses cdk.json with jq. If the config structure changes (e.g., helm.volcano.enabled moves), update the jq queries.

Testing Changes

After editing the script, do a dry run:

# Check syntax
bash -n demo/live_demo.sh

# Run with a quick pass (shorter pauses)
GCO_DEMO_FAST=1 bash demo/live_demo.sh

Troubleshooting

Problem Fix
jq: command not found Install jq: brew install jq or apt install jq
gco: command not found Install GCO CLI: pipx install -e . from repo root
Jobs stuck in Pending Check node provisioning: kubectl get nodes -w — GPU nodes take 60-90s
Script skips a scheduler you enabled Re-run gco stacks deploy-all -y after changing cdk.json
Colors not rendering Ensure your terminal supports ANSI colors. Try TERM=xterm-256color
FSx/Valkey section skipped Verify fsx_lustre.enabled / valkey.enabled is true in cdk.json