Tech Stack Options & Considerations

This document presents technology alternatives for consideration. These are SUGGESTIONS, not definitive choices. Each option includes trade-offs for the team to evaluate.

Decision Framework

When evaluating alternatives, consider:

---

Database Options

Option A: Prisma Postgres (Current Choice)

Managed PostgreSQL with integrated ORM, connection pooling, and edge caching.

Pros:

• Unified platform (DB + ORM + Caching)
• Serverless-optimized connection pooling
• Simple pricing ($49/mo Pro plan)
• Built-in edge caching (Prisma Accelerate, 300+ locations)
• Type-safe queries with Prisma Client
• Excellent developer experience
• No additional infrastructure to manage

Cons:

• Single vendor dependency
• Limited regional availability
• Newer platform (less battle-tested at scale)
• No built-in multi-region replication

Best For: MVP, startups, teams prioritizing simplicity

---

Option B: CockroachDB (Alternative Suggestion)

Distributed SQL database with built-in multi-region support.

Pros:

• Distributed SQL with ACID guarantees
• Multi-region by design (survives region failures)
• PostgreSQL wire-compatible
• Prevents "double-booking" across regions
• Automatic sharding and rebalancing
• Strong consistency across regions

Cons:

• Higher operational complexity
• Higher cost (~$300+/month for dedicated)
• Requires more expertise to tune
• Latency overhead for cross-region consistency
• Overkill for single-region deployment

Best For: Multi-region deployments, high-availability requirements

---

Recommendation

---

Caching Options

Option A: Prisma Accelerate (Current Choice)

Query-level caching built into Prisma Postgres.

Pros:

• Built into Prisma Postgres (no additional service)
• 300+ edge locations globally
• Query-level caching
• Automatic cache invalidation
• No additional infrastructure

Cons:

• Limited to query caching
• No pub/sub capability
• No session storage
• No custom data structures
• Cannot implement 10-minute inventory holds

Best For: Query caching, read-heavy workloads

---

Option B: Redis (Alternative Suggestion)

In-memory data store for caching, sessions, and real-time features.

Pros:

• Sub-millisecond latency
• Ideal for 10-minute inventory holds (TTL support)
• Session storage
• Pub/sub for real-time updates
• Rich data structures (sets, sorted sets, hashes)
• Widely supported, battle-tested

Cons:

• Additional service to manage
• Additional cost (~$50-100/month for managed)
• Requires connection management
• Memory-bound (can get expensive at scale)

Best For: Inventory holds, sessions, real-time features

---

Recommendation

┌─────────────────────────────────────────────────────────────────┐
│                                                                 │
│   Recommended: USE BOTH                                         │
│                                                                 │
│   Prisma Accelerate → Query caching                             │
│   Redis             → Inventory holds, sessions, real-time      │
│                                                                 │
│   They serve different purposes and complement each other.      │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

---

Messaging Options

Option A: EventBridge (Current Choice)

AWS-native event bus for event-driven architecture.

Pros:

• Native AWS integration
• Serverless (no infrastructure)
• Event filtering and routing
• Low cost at moderate volume
• Schema registry
• Easy to set up

Cons:

• Limited throughput (~10K events/sec)
• No event replay/reprocessing
• AWS-specific (vendor lock-in)
• Limited consumer patterns
• No stream processing

Best For: Event routing, serverless workflows, moderate volume

---

Option B: Apache Kafka (Alternative Suggestion)

Distributed event streaming platform.

Pros:

• Massive throughput (millions of events/sec)
• Event replay and reprocessing
• Multi-consumer patterns
• Stream processing capabilities
• Industry standard
• Portable across clouds

Cons:

• Significant operational overhead
• Higher cost (AWS MSK ~$200+/month minimum)
• Requires expertise to operate
• Overkill for current scale
• Complex configuration

Best For: High-volume streaming, event sourcing, complex processing

---

Recommendation

---

Identity Options

Option A: Cognito (Current Choice)

AWS-native identity and access management.

Pros:

• Native AWS integration
• Low cost (free tier generous)
• Built-in MFA
• Serverless
• JWT tokens out of the box
• Social login support

Cons:

• Limited customization
• Vendor lock-in to AWS
• Basic analytics and insights
• Limited branding options
• Clunky user migration

Best For: AWS-native apps, cost-sensitive projects

---

Option B: Auth0/Okta (Alternative Suggestion)

Enterprise identity platform with advanced features.

Pros:

• Superior developer experience
• Advanced security (breached password detection)
• Better social login support
• Cross-cloud portable
• Extensive customization
• Better analytics and insights
• Universal Login page

Cons:

• Significant cost ($3-5K/month at scale)
• External dependency (latency)
• Can be complex to configure
• Pricing scales with MAU

Best For: Enterprise features, multi-cloud, advanced security

---

Recommendation

---

Protocol Options

Option A: REST/JSON (Current Choice)

Standard HTTP REST APIs with JSON payloads.

Pros:

• Universal compatibility
• Easy to debug (human-readable)
• Extensive tooling
• No special client libraries needed
• Browser-friendly

Cons:

• Higher latency than binary protocols
• Larger payload sizes
• No built-in streaming
• No strong typing (without OpenAPI)

Best For: Public APIs, browser clients, aggregator integration

---

Option B: gRPC (Alternative Suggestion)

Binary protocol with strong typing and streaming.

Pros:

• Sub-millisecond latency
• Smaller payloads (Protocol Buffers)
• Strong typing with code generation
• Bi-directional streaming
• Better for service-to-service

Cons:

• Not browser-native (needs proxy)
• Harder to debug
• Learning curve
• Less tooling than REST

Best For: Internal microservices, high-performance paths

---

Recommendation

┌─────────────────────────────────────────────────────────────────┐
│                                                                 │
│   External APIs (NDC, aggregators)  → REST/JSON                 │
│   Internal services                 → Evaluate gRPC             │
│                                                                 │
│   Priority paths (inventory lock)   → Consider gRPC             │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

---

Cloud Provider

Current: AWS

No compelling reason to change. Benefits:

• Lambda for serverless compute
• API Gateway for API management
• EventBridge for events
• S3 for storage
• CloudFront for CDN
• Regional presence in South America

Alternative Consideration

Multi-cloud is not recommended at this stage:

• Adds complexity without clear benefit
• Higher operational cost
• Most vendors (Prisma, etc.) work well with AWS

---

Summary Matrix

---

Decision Log

---

Review Triggers

Revisit these decisions when:

• Multi-region deployment required → CockroachDB
• Event volume >10K/sec → Kafka
• Enterprise SSO required → Auth0/Okta
• Internal service latency critical → gRPC
• Inventory hold failures occur → Redis tuning