Yesterday in one word

One team from each region shouts one word that sums up Day 1. Then, the standings.

Open leaderboard

Four things that decide if they come back

Why p95, not average? If 95 people wait 0.2s and 5 people wait 30s, the average looks “fine” — those 5 are tweeting about you. 🐦

How much downtime per year is allowed?

Each extra nine costs roughly 10× more effort & money. Does a chai app need five nines? Does UPI?

How reliable systems are built

Monitoring vs Observability

The 4 Golden Signals (Google SRE)

💡 Remember “saturation” and “connection pools”… you'll need it in 2 minutes.

Six suspects. One killer. Maybe an accomplice.

Rules of the investigation 🕵️

Let's walk the evidence chain

1. 🟢 Postgres Pete — innocent. DB CPU flat at ~28%. His active connections dropped from ~42 to 5 at 08:59. He wasn't overloaded — he was starved.
2. 🟢 PayFast Priya — innocent. Latency flat ~205 ms. Failed traces never even reach her. Calls to her dropped.
3. 🟢 Redis Rani & Disk-Full Dinesh — innocent. Flat for days. Classic red herrings (real issues, wrong incident).
4. 🔴 Deploy Dave — THE KILLER. 08:58 deploy set maximumPoolSize=5 (was 50) — ticket FINOPS-482 “reduce DB cost”, load test skipped.
5. 🟠 Traffic Tara — THE ACCOMPLICE. 3× traffic at 09:00 exhausted 5 connections → requests wait 30 s → Connection is not available → checkout 504s.

What real SREs take away

Fix: roll back now. Prevent: load test in CI, canary deploys, SLO alerts on payment latency, pool-saturation alerts.

What each signal looks like

checkout.success_rate
08:59  99.1%
09:01  97.8%
09:03  38.2%

09:02:43 ERROR [payment]
HikariPool-1 - Connection
is not available, request
timed out after 30000ms

checkout   ███████████ 30.0s
 ├ cart    ▏ 19ms
 └ payment ██████████ 30.0s
    └ getConnection ███ 30.0s

Plus: events (deployments, config changes), profiles (code hotspots), real-user data. Standard for collecting them all: OpenTelemetry (CNCF).

Remember the kitchen order ticket? 🎫

Hunt for clues in a real environment

Answers and proof: ChaiCart Live · +10 per correct answer

Asking questions of your telemetry

// How many errors, and when did they spike?
fetch logs
| filter loglevel == "ERROR"
| makeTimeseries errors = count(), interval: 5m

// Which services are the slowest (p95)?
fetch spans
| summarize p95 = percentile(duration, 95), by: { service.name }
| sort p95 desc
| limit 10

// Is any host saturated?
timeseries cpu = avg(dt.host.cpu.usage), by: { dt.entity.host }

Observability = being able to ask new questions without shipping new code.

“SRE is what happens when you ask a software engineer to design an operations team.”
— Ben Treynor Sloss, Google

Making reliability measurable

Spend your 43 minutes wisely

Write the postmortem for “Who Killed Checkout?”

DevOps = culture + automation + measurement

Sequential delivery vs continuous flow

Open ChaiCart Live. Six virtual changes per round: Plan → Build → Test → Deploy.

The DORA metrics

Deploy Dave would have been caught by a canary + an SLO alert. 😉

From commit to cloud in minutes

# .github/workflows/deploy.yml  (GitHub Actions → Azure App Service)
on: { push: { branches: [ main ] } }
jobs:
  build-and-deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20 }
      - run: npm ci && npm test          # ← CI: fail fast
      - uses: azure/webapps-deploy@v3     # ← CD: ship it
        with:
          app-name: chaicart-demo
          publish-profile: ${{ secrets.AZURE_WEBAPP_PUBLISH_PROFILE }}

Show: push a change to the page title → pipeline turns green → refresh the site. Then push a failing test → pipeline blocks the deploy. 🛑

Provider secures the cloud. You secure what's in the cloud.

⚠️ Most cloud breaches are customer misconfigurations — public storage, leaked keys, over-privileged accounts, no MFA.

The cloud security starter pack 🔐

Growing without breaking

The cloud is cheap… until the bill arrives 🧾

FinOps = engineering + finance + business sharing responsibility for cloud spend. Also watch data egress and log/telemetry volume — silent bill killers.

The month-end bill arrives…

What runs behind every “Order placed” message

Almost all of these are now delivered as SaaS — that's cloud computing's biggest business impact.

One chai order, seven business systems

Integration is where the real work is

“Every company is becoming a software company”

Top trends to watch 🔮

Draw your cloud destiny

Each member chooses a career card in My workshop. Read your “destiny”. Accept it, or swap once with a teammate. Then fill in your row of the 90-day plan in your individual app plan.

What to learn — in order

Hot-seat questions 🔥

Final quiz

Same rules as yesterday: COO submits the team answer in ChaiCart Live. The facilitator locks each question and applies scores.

Open live quiz controls

Demo Day