Building Live Streaming Dashboards in Grafana

Multi-panel build · Variables · Assignment 3 guidance

Learning Objectives

• ✓ Extend the Lecture 14 transaction dashboard into a five-panel fraud-operations view
• ✓ Use dashboard variables to make charts interactive and filterable
• ✓ Apply dashboard design principles: hierarchy, colour meaning, panel arrangement
• ✓ Connect visualisation choices to the business questions being answered
• ✓ Distinguish historical event time from live ingestion time
• ✓ Produce a dashboard suitable for Assignment 3 presentation

Session Timeline

Time	Segment	Activity
0–10 min	Setup Check	Open the Lecture 14 Transaction Risk dashboard
10–20 min	Design Principles	What makes a dashboard useful vs pretty
20–55 min	Lab Part 1	Extend the transaction dashboard to five panels
55–70 min	Lab Part 2	Add a city variable and update panel queries
70–75 min	Interpretation	Turn dashboard evidence into a management action
75–85 min	Design Review	Student dashboards — what works, what doesn't
85–90 min	A3 Guidance	Assignment 3 requirements and tips

Section 1: Dashboard Design Principles

Section 2: Lab Part 1 — Fraud Operations Dashboard

Set the dashboard time range to 1 August–31 August 2025 so the Daily Transaction Volume panel includes the imported historical records.

Target Layout

┌───────────────────────────────────────────────────────┐
│ Fraud Operations  [City: All ▼]                       │
├───────────────────┬───────────────────────────────────┤
│ Flagged Rate      │ Flagged Transaction Value         │
│ (Stat)            │ (Stat)                            │
├───────────────────┴───────────────────────────────────┤
│ Daily Transaction Volume (Time Series)                 │
├───────────────────────────┬───────────────────────────┤
│ Flagged by Merchant       │ Highest-Value Flagged     │
│ (Bar Chart)               │ Transactions (Table)      │
└───────────────────────────────────────────────────────┘

Keep the fraud-rate Stat and review-queue Table from Lecture 14. Add the three panels below, then arrange all five into the target layout.

Panel 1 — Flagged Transaction Value (Stat)

SELECT SUM(amount) AS value
FROM transactions
WHERE is_fraud = 'true';

Settings: Visualisation = Stat | Unit = Indian Rupee | Decimals = 0 | Colour = red

Panel 2 — Daily Transaction Volume (Time Series)

SELECT
  $__timeGroupAlias(`timestamp`, '1d'),
  COUNT(*) AS value
FROM transactions
WHERE $__timeFilter(`timestamp`)
GROUP BY 1
ORDER BY 1;

Settings: Visualisation = Time Series | Unit = count | Line width = 2 | Fill opacity = 10

Panel 3 — Flagged Transactions by Merchant (Bar Chart)

SELECT merchant, COUNT(*) AS value
FROM transactions
WHERE is_fraud = 'true'
GROUP BY merchant
ORDER BY value DESC
LIMIT 10;

Settings: Visualisation = Bar Chart | Category = merchant | Value = count | Orientation = horizontal

Panel 4 — Fraud Rate (from Lecture 14)

Panel 5 — Review Queue (from Lecture 14)

Section 3: Lab Part 2 — Variables & Interactivity

Adding a City Dropdown

Variables make one dashboard reusable across locations instead of creating a separate dashboard for every city.

1. Dashboard settings (gear icon) → Variables → New variable
2. Name: city | Type: Custom | Values: Mumbai, Delhi, Bengaluru, Hyderabad, Chennai, Kolkata, Pune, Ahmedabad
3. Enable Include All option and set the custom all value to .*.
4. Add AND city REGEXP '^${city:raw}$' to queries that already have a WHERE clause.
5. For queries without WHERE, add WHERE city REGEXP '^${city:raw}$' before GROUP BY.

The dropdown now filters all panels simultaneously. Selecting All uses the .* regular expression and keeps every city.

Saving & Sharing

1. Save icon → name: "SDA — Fraud Operations"
2. Share → Snapshot → generates a public link (no Grafana login required to view)

Section 4: Live Dashboard Mode

The dashboard above analyses a completed historical dataset. To watch it change while Kafka is running, use the isolated Lecture 16 Extension files and database. One consumer writes every transaction to sda_course_l16.transactions; the other writes flagged events to sda_course_l16.fraud_alerts.

transaction_producer_16.py
          |
          v
  transactions-topic-l16
       |          |
       v          v
consumer_store_  consumer_alerts_
mysql_16.py      mysql_16.py
       |          |
       v          v
transactions     fraud_alerts
       \          /
        v        v
          Grafana

1. Set the dashboard range to Last 5 minutes.
2. Set dashboard refresh to 5s.
3. Use ingested_at for live transaction panels, not the simulated timestamp.
4. Use detected_at for the live alert queue.

SELECT
  $__timeGroupAlias(ingested_at, '5s'),
  COUNT(*) AS value
FROM transactions
WHERE $__timeFilter(ingested_at)
GROUP BY 1
ORDER BY 1;

Section 5: Management Interpretation

Use the completed dashboard to prepare a 60-second fraud-operations briefing:

1. State the risk: What is the flagged rate and flagged transaction value?
2. Locate the workload: Which days, merchants, and cities dominate activity?
3. Avoid overclaiming: Explain why flagged does not mean confirmed fraud.
4. Recommend action: Name one investigation, control, or threshold change.

Section 6: Design Review Checklist

For each dashboard, evaluate:

Question	What to Look For
Does it answer a clear business question?	"Who is the user? What decision are they making?" — one sentence
Is colour usage consistent?	Red = alert/bad, Green = normal. No decorative red on neutral data.
Is the refresh rate appropriate?	30s is fine for demos; 5s for live risk monitoring
Are axes labelled with units?	Always: ₹, count, %, minutes — never raw numbers without context
Is the time range visible?	Users should always know what window of data they are looking at

Assignment 3 Guidance

Due: 1 October | 15 marks

Create a dashboard for the analysis of your consumed data (the domain from A1, streamed in A2).

Requirement	Detail
Minimum 4 panels	At least 1 time series, 1 stat/KPI, 1 table
Real data	Connected to the course MySQL transaction dataset or another supported datasource
1 variable/filter	Makes the dashboard interactive
Colour thresholds	Reflect actual business logic — not default colours
Explanation	200-word write-up: what business question each panel answers

Submission: Export dashboard JSON (Dashboard → Share → Export JSON) + screenshots. Upload to GitHub with tag [A3][SDA-1], [A3][SDA-2], or [A3][SDA-G].

Key Takeaways