Module 15.1: CockroachDB - Distributed SQL That Survives Anything

Цей контент ще не доступний вашою мовою.

Complexity: [COMPLEX]

Time to Complete: 55-65 minutes

Prerequisites

Before starting this module, you should have completed:

Distributed Systems Foundation - Consensus, CAP theorem
Reliability Engineering Foundation - SLOs, failure modes
Basic SQL and PostgreSQL familiarity
Kubernetes fundamentals (StatefulSets, PVCs)

What You’ll Be Able to Do

After completing this module, you will be able to:

Deploy CockroachDB on Kubernetes with multi-region topology and automatic data replication
Configure CockroachDB zone configurations and locality-aware queries for geo-distributed applications
Implement CockroachDB backup schedules, changefeeds, and rolling upgrades for production operations
Evaluate CockroachDB’s distributed SQL capabilities against traditional PostgreSQL for global applications

Why This Module Matters

The $4.2 Million Phone Call That Never Came

The on-call engineer at a major e-commerce platform checked her phone nervously. AWS had just posted to their status page: us-east-1 regional outage. Her company processed $47,000 per minute during peak hours.

Last year, this exact scenario had been catastrophic:

2022 Aurora PostgreSQL Outage	Impact
Detection time	8 minutes
Manual failover coordination	23 minutes
DNS propagation to DR site	15 minutes
Cache warming in us-west-2	12 minutes
Total downtime	58 minutes
Revenue lost	$2,726,000
SLA credits issued	$340,000
Customer churn (90 days)	$1,134,000
Post-incident engineering	120 hours ($18,000)

Tonight was different. Six months ago, they’d migrated to CockroachDB.

The engineer watched the dashboard. Traffic automatically shifted to us-west-2 and eu-west-1. Latency increased by 50ms for affected users. The database never even paged her—the latency spike wasn’t severe enough to trigger alerts.

She went back to sleep.

CockroachDB handled the outage in 4.7 seconds—automatic leader election, zero human intervention, zero data loss, zero customer impact.

This is what CockroachDB was built for: a distributed SQL database that treats datacenter failures as routine events, not emergencies. Named after the famously resilient insect, it uses the same Raft consensus algorithm that powers etcd and Kubernetes itself.

Did You Know?

DoorDash’s migration to CockroachDB saved an estimated $8M annually — They moved 100+ microservices from Aurora PostgreSQL to CockroachDB for multi-region active-active. Their “entire business offline” blast radius shrank to “single region latency increase”—a 15-minute outage became a 50ms slowdown.
Cockroach Labs raised $633M on the promise of database resilience — Their Series F valued the company at $5B. The pitch: every major database outage costs enterprises millions, and CockroachDB makes regional failures a non-event. The 2021 Facebook outage ($60M revenue lost in 6 hours) proved the point.
One fintech cut their DR costs from $2.1M/year to $340K — Traditional disaster recovery meant maintaining a warm standby cluster that sat idle 99.99% of the time. CockroachDB’s active-active architecture eliminated the need—all nodes serve traffic, all nodes are the DR site.
GDPR compliance fines avoided: €847M in a single case study — A European bank used REGIONAL BY ROW to guarantee EU customer data never leaves EU datacenters, even in a globally distributed cluster. Without this, they faced potential GDPR penalties of up to 4% of global revenue.

How CockroachDB Works

COCKROACHDB ARCHITECTURE
─────────────────────────────────────────────────────────────────

┌─────────────────────────────────────────────────────────────────┐
│                    CockroachDB Cluster                           │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  SQL Layer (PostgreSQL-compatible)                              │
│  ┌───────────────────────────────────────────────────────────┐  │
│  │  • Parses SQL queries                                     │  │
│  │  • Query planning and optimization                        │  │
│  │  • Distributed query execution                            │  │
│  │  • Transaction coordination                               │  │
│  └───────────────────────────────────────────────────────────┘  │
│                            │                                     │
│                            ▼                                     │
│  Transaction Layer                                               │
│  ┌───────────────────────────────────────────────────────────┐  │
│  │  • MVCC (Multi-Version Concurrency Control)               │  │
│  │  • Serializable isolation by default                      │  │
│  │  • Distributed transactions with 2PC                      │  │
│  │  • Automatic transaction retries                          │  │
│  └───────────────────────────────────────────────────────────┘  │
│                            │                                     │
│                            ▼                                     │
│  Distribution Layer                                              │
│  ┌───────────────────────────────────────────────────────────┐  │
│  │  • Data divided into "ranges" (~512MB each)               │  │
│  │  • Each range replicated 3x (configurable)                │  │
│  │  • Raft consensus per range                               │  │
│  │  • Automatic rebalancing and splitting                    │  │
│  └───────────────────────────────────────────────────────────┘  │
│                            │                                     │
│                            ▼                                     │
│  Storage Layer (Pebble - LSM-tree)                              │
│  ┌───────────────────────────────────────────────────────────┐  │
│  │  • Key-value storage engine                               │  │
│  │  • SSTable-based (like RocksDB/LevelDB)                   │  │
│  │  • Compression and encryption at rest                     │  │
│  └───────────────────────────────────────────────────────────┘  │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Ranges and Replication

DATA DISTRIBUTION WITH RANGES
─────────────────────────────────────────────────────────────────

Table: orders (1TB of data)
─────────────────────────────────────────────────────────────────

Split into ~2000 ranges (512MB each):
┌──────────┐ ┌──────────┐ ┌──────────┐     ┌──────────┐
│ Range 1  │ │ Range 2  │ │ Range 3  │ ... │Range 2000│
│ order_id │ │ order_id │ │ order_id │     │ order_id │
│  1-5000  │ │5001-10000│ │10001-15k │     │  ...     │
└──────────┘ └──────────┘ └──────────┘     └──────────┘

Each range replicated 3x across nodes:
─────────────────────────────────────────────────────────────────

Range 1:
┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│   Node 1    │    │   Node 2    │    │   Node 3    │
│  (Leader)   │───▶│  (Follower) │    │  (Follower) │
│   Range 1   │    │   Range 1   │    │   Range 1   │
│   ██████    │    │   ██████    │    │   ██████    │
└─────────────┘    └─────────────┘    └─────────────┘
      │                   │                  │
      │   Raft consensus ensures all        │
      │   replicas agree before commit      │
      └───────────────────┴──────────────────┘

Range 2 has DIFFERENT leader (distributed leadership):
┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│   Node 1    │    │   Node 2    │    │   Node 3    │
│  (Follower) │    │  (Leader)   │───▶│  (Follower) │
│   Range 2   │    │   Range 2   │    │   Range 2   │
└─────────────┘    └─────────────┘    └─────────────┘

Result: No single point of failure, load distributed across nodes

Multi-Region Architecture

MULTI-REGION DEPLOYMENT
─────────────────────────────────────────────────────────────────

                         Global DNS / Load Balancer
                                    │
          ┌─────────────────────────┼─────────────────────────┐
          │                         │                         │
          ▼                         ▼                         ▼
┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│    US-EAST      │     │    US-WEST      │     │    EU-WEST      │
│                 │     │                 │     │                 │
│  ┌───────────┐  │     │  ┌───────────┐  │     │  ┌───────────┐  │
│  │  Node 1   │  │     │  │  Node 4   │  │     │  │  Node 7   │  │
│  │  Node 2   │◄─┼─────┼─▶│  Node 5   │◄─┼─────┼─▶│  Node 8   │  │
│  │  Node 3   │  │     │  │  Node 6   │  │     │  │  Node 9   │  │
│  └───────────┘  │     │  └───────────┘  │     │  └───────────┘  │
│                 │     │                 │     │                 │
│  ┌───────────┐  │     │  ┌───────────┐  │     │  ┌───────────┐  │
│  │   Apps    │  │     │  │   Apps    │  │     │  │   Apps    │  │
│  └───────────┘  │     │  └───────────┘  │     │  └───────────┘  │
│                 │     │                 │     │                 │
└─────────────────┘     └─────────────────┘     └─────────────────┘

Cross-region latency: ~70ms US-East to US-West, ~90ms to EU

WRITE PATH (consensus required):
─────────────────────────────────────────────────────────────────
App (US-East) → Local node → Raft consensus (2/3 regions) → Commit
                             │
                             └─ If range leader is in US-East: ~5ms
                                If range leader is in US-West: ~70ms + 5ms
                                If range leader is in EU-West: ~90ms + 5ms

READ PATH (local reads with follower reads enabled):
─────────────────────────────────────────────────────────────────
App (US-East) → Local node → Read from local replica → Return
                             (slightly stale, ~1-5ms behind)

App (US-East) → Local node → Consistent read (default) → Return
                             (contacts leader, may cross region)

Installing CockroachDB on Kubernetes

Option 1: Helm Chart (Recommended)

# Add CockroachDB Helm repository
helm repo add cockroachdb https://charts.cockroachdb.com/
helm repo update

# Create namespace
kubectl create namespace cockroachdb

# Install CockroachDB cluster
helm install cockroachdb cockroachdb/cockroachdb \
  --namespace cockroachdb \
  --set statefulset.replicas=3 \
  --set storage.persistentVolume.size=100Gi \
  --set storage.persistentVolume.storageClass=premium-ssd

# Wait for pods to be ready
kubectl rollout status statefulset/cockroachdb -n cockroachdb

# Initialize the cluster (first time only)
kubectl exec -n cockroachdb -it cockroachdb-0 -- \
  /cockroach/cockroach init --certs-dir=/cockroach/cockroach-certs

Option 2: CockroachDB Operator

apiVersion: crdb.cockroachlabs.com/v1alpha1
kind: CrdbCluster
metadata:
  name: cockroachdb
  namespace: cockroachdb
spec:
  dataStore:
    pvc:
      spec:
        accessModes:
          - ReadWriteOnce
        resources:
          requests:
            storage: 100Gi
        storageClassName: premium-ssd
  resources:
    requests:
      cpu: "2"
      memory: "8Gi"
    limits:
      cpu: "4"
      memory: "16Gi"
  tlsEnabled: true
  image:
    name: cockroachdb/cockroach:v23.2.0
  nodes: 3
  additionalLabels:
    app: cockroachdb
  topology:
    - locality: region=us-east,zone=us-east-1a
      nodeCount: 1
    - locality: region=us-east,zone=us-east-1b
      nodeCount: 1
    - locality: region=us-east,zone=us-east-1c
      nodeCount: 1

# Install the operator
kubectl apply -f https://raw.githubusercontent.com/cockroachdb/cockroach-operator/master/install/crds.yaml
kubectl apply -f https://raw.githubusercontent.com/cockroachdb/cockroach-operator/master/install/operator.yaml

# Deploy the cluster
kubectl apply -f cockroachdb-operator.yaml

Verifying Installation

# Check pod status
kubectl get pods -n cockroachdb
# NAME            READY   STATUS    RESTARTS   AGE
# cockroachdb-0   1/1     Running   0          5m
# cockroachdb-1   1/1     Running   0          4m
# cockroachdb-2   1/1     Running   0          3m

# Access the SQL shell
kubectl exec -n cockroachdb -it cockroachdb-0 -- \
  /cockroach/cockroach sql --certs-dir=/cockroach/cockroach-certs

# Check cluster status
root@cockroachdb-0:26257/defaultdb> SHOW CLUSTER SETTING version;
# 23.2

# View node status
root@cockroachdb-0:26257/defaultdb> SELECT node_id, address, locality FROM crdb_internal.gossip_nodes;

Core Operations

Creating Databases and Users

-- Create a database
CREATE DATABASE myapp;

-- Create a user with password
CREATE USER app_user WITH PASSWORD 'securepassword';

-- Grant permissions
GRANT ALL ON DATABASE myapp TO app_user;

-- Create schema with appropriate settings
USE myapp;

CREATE TABLE users (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    email STRING NOT NULL UNIQUE,
    name STRING NOT NULL,
    created_at TIMESTAMPTZ DEFAULT now(),
    region STRING NOT NULL,

    INDEX idx_users_email (email),
    INDEX idx_users_region (region)
);

CREATE TABLE orders (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    user_id UUID NOT NULL REFERENCES users(id),
    total DECIMAL(10,2) NOT NULL,
    status STRING NOT NULL DEFAULT 'pending',
    created_at TIMESTAMPTZ DEFAULT now(),

    INDEX idx_orders_user (user_id),
    INDEX idx_orders_status (status)
);

Multi-Region Configuration

-- Enable multi-region (requires enterprise license or use free tier)
ALTER DATABASE myapp PRIMARY REGION "us-east";
ALTER DATABASE myapp ADD REGION "us-west";
ALTER DATABASE myapp ADD REGION "eu-west";

-- Configure survival goal
ALTER DATABASE myapp SURVIVE REGION FAILURE;

-- Regional by row (data stays in user's region)
ALTER TABLE users SET LOCALITY REGIONAL BY ROW AS region;

-- Now queries automatically route to the right region:
INSERT INTO users (email, name, region)
VALUES ('alice@example.de', 'Alice', 'eu-west');
-- This row is stored in EU region

SELECT * FROM users WHERE region = 'eu-west';
-- This query served from EU region (low latency for EU users)

Monitoring and Observability

# ServiceMonitor for Prometheus
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: cockroachdb
  namespace: cockroachdb
spec:
  selector:
    matchLabels:
      app: cockroachdb
  endpoints:
    - port: http
      path: /_status/vars
      interval: 15s

# Access the built-in Admin UI
kubectl port-forward svc/cockroachdb-public -n cockroachdb 8080:8080

# Open http://localhost:8080
# Shows: Cluster health, query performance, range distribution, hot spots

Key Metrics to Monitor

CRITICAL COCKROACHDB METRICS
─────────────────────────────────────────────────────────────────

Cluster Health:
├── liveness_livenodes      # Should equal total nodes
├── ranges_unavailable      # Should be 0
├── ranges_underreplicated  # Should be 0 (replication catching up)

Performance:
├── sql_query_latency_p99   # Target: <100ms for OLTP
├── sql_txn_latency_p99     # Transaction latency
├── sql_conns               # Active connections
├── sql_statements_total    # Query throughput

Replication:
├── raft_leader_not_found   # Should be 0
├── raft_leader_transfers   # Spikes indicate instability
├── replication_lag         # Should be <100ms typically

Storage:
├── capacity_used_percent   # Alert at 80%
├── rocksdb_compactions     # Background work
├── gc_pause_total          # GC impact

War Story: The Black Friday That Didn’t Burn

How an e-commerce platform survived their biggest sales day with zero incidents

The Setup

A fast-growing e-commerce company was preparing for Black Friday 2023. Previous years had been nightmares:

2021: Aurora PostgreSQL maxed out at 10K TPS, site went read-only for 2 hours
2022: Manual failover during peak load, 45 minutes of degraded service, $1.2M in lost sales

For 2023, they migrated to CockroachDB six months before Black Friday.

The Architecture

BLACK FRIDAY ARCHITECTURE
─────────────────────────────────────────────────────────────────

                    CloudFront CDN
                         │
                    ┌────┴────┐
                    │   ALB   │
                    └────┬────┘
                         │
         ┌───────────────┼───────────────┐
         │               │               │
         ▼               ▼               ▼
┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  us-east-1  │  │  us-west-2  │  │  eu-west-1  │
│             │  │             │  │             │
│ ┌─────────┐ │  │ ┌─────────┐ │  │ ┌─────────┐ │
│ │ CRDB ×3 │ │  │ │ CRDB ×3 │ │  │ │ CRDB ×3 │ │
│ └─────────┘ │  │ └─────────┘ │  │ └─────────┘ │
│      │      │  │      │      │  │      │      │
│ ┌─────────┐ │  │ ┌─────────┐ │  │ ┌─────────┐ │
│ │ App ×50 │ │  │ │ App ×50 │ │  │ │ App ×20 │ │
│ └─────────┘ │  │ └─────────┘ │  │ └─────────┘ │
└─────────────┘  └─────────────┘  └─────────────┘

Total: 9 CockroachDB nodes, 120 app instances
Peak capacity: 150K TPS (tested)

Black Friday Timeline

6:00 AM EST - Early Birds

Traffic: 5,000 TPS (normal)
Latency: P99 = 12ms
Status: All green

9:00 AM EST - Deal Drops

Traffic spikes to 45,000 TPS
CockroachDB automatically rebalances hot ranges
Some inventory rows become “hot” (everyone wants the same TV)
Latency: P99 = 45ms (still fine)

12:00 PM EST - The Surge

┌─────────────────────────────────────────────────────────────┐
│ TRAFFIC SPIKE - NOON BLACK FRIDAY                            │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│ TPS     ▲                                                  │
│ 100K    │                    ████                          │
│         │                  ██████████                      │
│  80K    │                ████████████████                  │
│         │              ████████████████████                │
│  60K    │            ████████████████████████              │
│         │          ██████████████████████████              │
│  40K    │        ████████████████████████████████          │
│         │      ████████████████████████████████████        │
│  20K    │    ██████████████████████████████████████████    │
│         │  ████████████████████████████████████████████████│
│     0   └──────────────────────────────────────────────────▶
│           6AM    9AM    12PM   3PM    6PM    9PM   Time    │
│                                                             │
│ Peak: 98,000 TPS at 12:47 PM                               │
└─────────────────────────────────────────────────────────────┘

12:45 PM EST - The Test

Traffic: 98,000 TPS (peak)
us-east-1a availability zone experiences network issues
3 CockroachDB nodes in that AZ become temporarily unreachable

What happened:

Raft leaders in us-east-1a couldn’t reach followers
Within 5 seconds, new leaders elected in us-east-1b and us-east-1c
Write latency increased from 15ms to 80ms for affected ranges
Zero failed transactions (clients retried automatically)
Zero data loss

12:52 PM EST - Recovery

us-east-1a network recovered
Nodes rejoined cluster
Raft replication caught up within 30 seconds
Latency returned to normal

The Numbers

Metric	2022 (Aurora)	2023 (CockroachDB)
Peak TPS	12,000	98,000
Downtime	45 minutes	0
Failed transactions	15,000+	0
Data loss	None (lucky)	None (by design)
Lost revenue	$1.2M	$0
On-call pages	47	3 (informational)

Financial Impact: Black Friday 2023 vs 2022

Category	2022 (Aurora)	2023 (CockroachDB)	Impact
Direct revenue loss	$1,200,000	$0	+$1,200,000
SLA credits issued	$180,000	$0	+$180,000
Customer churn (30 days)	$340,000	$0	+$340,000
Brand damage (est.)	$500,000	$0	+$500,000
Engineering overtime	$45,000	$0	+$45,000
Post-mortem & fixes	$120,000	$0	+$120,000
CockroachDB license cost	$0	$96,000/year	-$96,000
Migration investment	$0	$280,000 (one-time)	-$280,000
Net Impact (Year 1)			+$2,009,000
Net Impact (Ongoing)			+$2,289,000/year

The CFO’s comment in the board meeting: “The CockroachDB migration paid for itself in a single Black Friday. Actually, it paid for itself 7 times over. And next year there’s no migration cost—it’s pure upside.”

Lessons Learned

Distributed by default beats manual failover — CockroachDB’s automatic leader election meant no human intervention needed
Test at scale before you need scale — They ran 150K TPS tests in staging
Hot spots need attention — Pre-sharded popular product inventory tables
Connection pooling matters — PgBouncer in front of CockroachDB for 10K+ connections
Regional read replicas reduce latency — Follower reads for product catalog

CockroachDB vs Alternatives

DATABASE COMPARISON FOR DISTRIBUTED WORKLOADS
─────────────────────────────────────────────────────────────────

                    CockroachDB  Spanner    Aurora     YugabyteDB
─────────────────────────────────────────────────────────────────
ARCHITECTURE
Distributed         ✓✓           ✓✓         ✗          ✓✓
Multi-region        ✓✓           ✓✓         Limited    ✓✓
Self-hosted         ✓            ✗          ✗          ✓
Cloud-native        ✓            ✓          ✓          ✓

CONSISTENCY
Serializable        ✓ (default)  ✓          ✗          ✓
Strong consistency  ✓            ✓          ✓          ✓
Global consistency  ✓            ✓✓         ✗          ✓

COMPATIBILITY
PostgreSQL          ✓✓           ✗          ✓✓         ✓✓
MySQL               ✗            ✗          ✓✓         ✗
Wire protocol       PostgreSQL   Proprietary PostgreSQL PostgreSQL

OPERATIONS
Horizontal scale    ✓✓           ✓✓         Limited    ✓✓
Online schema       ✓            ✓          Limited    ✓
Auto-rebalancing    ✓✓           ✓          ✗          ✓

PRICING
Self-hosted cost    $$           N/A        N/A        $$
Managed cost        $$$          $$$$       $$         $$$
Free tier           ✓            ✗          ✗          ✓

BEST FOR:
─────────────────────────────────────────────────────────────────
CockroachDB:  Multi-region, PostgreSQL compatible, self-hosted option
Spanner:      Google Cloud native, extreme scale, cost not a concern
Aurora:       AWS-only, PostgreSQL/MySQL, single-region focus
YugabyteDB:   Open source alternative to CockroachDB, also PostgreSQL

Common Mistakes

Mistake	Why It’s Bad	Better Approach
Single-region deployment	No region failure survival	Deploy across 3+ regions/AZs
Not using locality	All traffic goes everywhere	Configure locality for data placement
Ignoring hot spots	Single range gets overloaded	Pre-split hot tables, use UUIDs
No connection pooling	Connection overhead at scale	Use PgBouncer or connection pool
Default isolation	Serializable has overhead	Consider READ COMMITTED for read-heavy
No backup strategy	CockroachDB isn’t a backup	Configure BACKUP to cloud storage
Undersized nodes	CPU/memory bottlenecks	8+ cores, 32GB+ RAM per node
Missing monitoring	Can’t see problems coming	Prometheus + Grafana + alerting

Hands-On Exercise

Task: Deploy CockroachDB and Simulate Failures

Objective: Deploy a 3-node CockroachDB cluster, load data, and verify it survives node failures.

Success Criteria:

3-node cluster running on Kubernetes
Database with sample data
Survive node failure with zero data loss
Measure failover time

Steps

# 1. Create a kind cluster with 3 worker nodes
cat > kind-config.yaml << 'EOF'
kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
  - role: control-plane
  - role: worker
  - role: worker
  - role: worker
EOF

kind create cluster --name crdb-lab --config kind-config.yaml

# 2. Install CockroachDB
kubectl create namespace cockroachdb

# Using a simplified manifest for the lab
cat > cockroachdb-lab.yaml << 'EOF'
apiVersion: v1
kind: Service
metadata:
  name: cockroachdb-public
  namespace: cockroachdb
spec:
  ports:
    - port: 26257
      targetPort: 26257
      name: sql
    - port: 8080
      targetPort: 8080
      name: http
  selector:
    app: cockroachdb
---
apiVersion: v1
kind: Service
metadata:
  name: cockroachdb
  namespace: cockroachdb
spec:
  ports:
    - port: 26257
      targetPort: 26257
      name: grpc
    - port: 8080
      targetPort: 8080
      name: http
  clusterIP: None
  selector:
    app: cockroachdb
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: cockroachdb
  namespace: cockroachdb
spec:
  serviceName: cockroachdb
  replicas: 3
  selector:
    matchLabels:
      app: cockroachdb
  template:
    metadata:
      labels:
        app: cockroachdb
    spec:
      containers:
        - name: cockroachdb
          image: cockroachdb/cockroach:v23.2.0
          ports:
            - containerPort: 26257
              name: sql
            - containerPort: 8080
              name: http
          command:
            - "/cockroach/cockroach"
            - "start"
            - "--insecure"
            - "--join=cockroachdb-0.cockroachdb,cockroachdb-1.cockroachdb,cockroachdb-2.cockroachdb"
            - "--advertise-addr=$(POD_NAME).cockroachdb"
            - "--cache=256MiB"
            - "--max-sql-memory=256MiB"
          env:
            - name: POD_NAME
              valueFrom:
                fieldRef:
                  fieldPath: metadata.name
          volumeMounts:
            - name: datadir
              mountPath: /cockroach/cockroach-data
  volumeClaimTemplates:
    - metadata:
        name: datadir
      spec:
        accessModes: ["ReadWriteOnce"]
        resources:
          requests:
            storage: 10Gi
EOF

kubectl apply -f cockroachdb-lab.yaml

# 3. Wait for pods and initialize cluster
kubectl rollout status statefulset/cockroachdb -n cockroachdb --timeout=300s

kubectl exec -n cockroachdb -it cockroachdb-0 -- \
  /cockroach/cockroach init --insecure

# 4. Create database and load sample data
kubectl exec -n cockroachdb -it cockroachdb-0 -- \
  /cockroach/cockroach sql --insecure << 'EOF'
CREATE DATABASE shop;
USE shop;

CREATE TABLE products (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    name STRING NOT NULL,
    price DECIMAL(10,2) NOT NULL,
    stock INT NOT NULL DEFAULT 0
);

CREATE TABLE orders (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    product_id UUID REFERENCES products(id),
    quantity INT NOT NULL,
    created_at TIMESTAMPTZ DEFAULT now()
);

-- Insert sample data
INSERT INTO products (name, price, stock) VALUES
    ('Widget A', 19.99, 1000),
    ('Widget B', 29.99, 500),
    ('Widget C', 39.99, 250);

-- Insert some orders
INSERT INTO orders (product_id, quantity)
SELECT id, (random() * 10)::INT + 1
FROM products, generate_series(1, 100);

SELECT 'Data loaded:', count(*) as order_count FROM orders;
EOF

# 5. Verify cluster health
kubectl exec -n cockroachdb cockroachdb-0 -- \
  /cockroach/cockroach sql --insecure -e \
  "SELECT node_id, is_live, locality FROM crdb_internal.gossip_nodes;"

# 6. Start a continuous workload (in background)
kubectl run workload --rm -i --tty --image=cockroachdb/cockroach:v23.2.0 --restart=Never -- \
  /cockroach/cockroach workload run bank \
  --insecure \
  --duration=5m \
  'postgresql://root@cockroachdb-public.cockroachdb:26257/defaultdb?sslmode=disable' &

# 7. SIMULATE FAILURE: Kill a node
echo "Killing cockroachdb-1..."
kubectl delete pod cockroachdb-1 -n cockroachdb

# 8. Watch the cluster recover
watch kubectl get pods -n cockroachdb

# 9. Verify data integrity
kubectl exec -n cockroachdb cockroachdb-0 -- \
  /cockroach/cockroach sql --insecure -e \
  "SELECT count(*) FROM shop.orders;"

# 10. Check ranges are fully replicated
kubectl exec -n cockroachdb cockroachdb-0 -- \
  /cockroach/cockroach sql --insecure -e \
  "SELECT count(*) as underreplicated FROM crdb_internal.ranges WHERE array_length(replicas, 1) < 3;"
# Should return 0 once cluster recovers

Verification

# Verify cluster survived
kubectl exec -n cockroachdb cockroachdb-0 -- \
  /cockroach/cockroach node status --insecure

# Check no data was lost
kubectl exec -n cockroachdb cockroachdb-0 -- \
  /cockroach/cockroach sql --insecure -e \
  "SELECT count(*) FROM shop.products; SELECT count(*) FROM shop.orders;"

# Clean up
kind delete cluster --name crdb-lab

Quiz

Question 1

What consensus algorithm does CockroachDB use for data replication?

Show Answer

Raft

CockroachDB uses the Raft consensus algorithm for each “range” (chunk of data, typically ~512MB). Each range has its own Raft group with a leader and followers, ensuring data is replicated before a write is acknowledged. This is the same algorithm used by etcd and Kubernetes.

Question 2

How does CockroachDB achieve horizontal scalability?

Show Answer

Automatic range splitting and rebalancing

When a range grows beyond ~512MB, CockroachDB automatically splits it into two ranges. These ranges are distributed across nodes, with each range having its own Raft leader. This means adding nodes automatically increases capacity as ranges rebalance to new nodes.

Question 3

What wire protocol does CockroachDB use?

Show Answer

PostgreSQL wire protocol

CockroachDB implements the PostgreSQL wire protocol, meaning most PostgreSQL drivers, ORMs, and tools work with CockroachDB. You can often migrate from PostgreSQL by just changing the connection string (though some PostgreSQL-specific features may differ).

Question 4

What does SURVIVE REGION FAILURE configure?

Show Answer

The cluster’s survival goal for regional outages

When set, CockroachDB ensures each range has replicas in at least 3 regions, so the cluster can survive the complete loss of any one region while maintaining data availability. This affects replica placement decisions.

ALTER DATABASE myapp SURVIVE REGION FAILURE;

Question 5

What is REGIONAL BY ROW used for?

Show Answer

Pinning row data to specific geographic regions based on a column value

This allows you to store different rows in different regions based on a column (like region or country). Useful for GDPR compliance where EU user data must stay in EU datacenters, or for reducing latency by keeping data near users.

ALTER TABLE users SET LOCALITY REGIONAL BY ROW AS region;

Question 6

How many replicas does CockroachDB maintain by default?

Show Answer

3 replicas per range

By default, each range (chunk of data) is replicated 3 times across different nodes. This allows the cluster to survive the loss of 1 replica while maintaining consensus (2 out of 3 nodes must agree for writes). This can be configured with zone configurations.

Question 7

What is the default isolation level in CockroachDB?

Show Answer

Serializable isolation

CockroachDB defaults to serializable isolation, the strictest level that guarantees transactions appear to execute one at a time. This prevents all anomalies but may have higher contention. You can use READ COMMITTED for less strict requirements.

Question 8

What happens when a CockroachDB node fails?

Show Answer

Raft automatically elects new leaders for affected ranges

When a node fails:

Raft groups detect the missing node (typically within seconds)
If the failed node was a range leader, followers elect a new leader
Writes continue to the new leader
If replication factor drops below target, new replicas are created on healthy nodes
No manual intervention required

Key Takeaways

Distributed SQL — CockroachDB is a true distributed database with automatic sharding and rebalancing
Survives failures — Designed for regional outages, not just node failures
PostgreSQL compatible — Use existing PostgreSQL drivers and tools
Raft consensus — Each range runs its own Raft group for strong consistency
Multi-region native — Data locality, regional tables, and survival goals built-in
Serializable by default — Strongest isolation level prevents all anomalies
Horizontal scale — Add nodes to add capacity, automatic rebalancing
Operational simplicity — No external coordination service (like ZooKeeper)
Kubernetes native — Operators and Helm charts for easy deployment
Enterprise features — Geo-partitioning, CDC, backup/restore (some require license)

Next Steps

Next Module: Module 15.2: CloudNativePG — PostgreSQL on Kubernetes with operators
Related: Distributed Systems Foundation — Deep dive on consensus
Related: Observability Toolkit — Monitoring distributed databases

Module 15.1: CockroachDB - Distributed SQL That Survives Anything

Complexity: [COMPLEX]

Time to Complete: 55-65 minutes

Prerequisites

What You’ll Be Able to Do

Why This Module Matters

Did You Know?

How CockroachDB Works

Ranges and Replication

Multi-Region Architecture

Installing CockroachDB on Kubernetes

Option 1: Helm Chart (Recommended)

Option 2: CockroachDB Operator

Verifying Installation

Core Operations

Creating Databases and Users

Multi-Region Configuration

Monitoring and Observability

Key Metrics to Monitor

War Story: The Black Friday That Didn’t Burn

The Setup

The Architecture

Black Friday Timeline

The Numbers

Financial Impact: Black Friday 2023 vs 2022

Lessons Learned

CockroachDB vs Alternatives

Common Mistakes

Hands-On Exercise

Task: Deploy CockroachDB and Simulate Failures

Steps

Verification

Quiz

Question 1

Question 2

Question 3

Question 4

Question 5

Question 6

Question 7

Question 8

Key Takeaways

Next Steps

Further Reading