Serverless Databases FAQ & Answers

Neon (PostgreSQL): Best cost-effectiveness for variable workloads, usage-based pricing $5/month minimum (Launch plan), $0.106/CU-hour compute + $0.35/GB-month storage, automatic scale-to-zero after 5 min inactivity, copy-on-write branching (instantaneous, zero data duplication), free tier with 30 projects and 100 CU-hours per project. PlanetScale (PostgreSQL + MySQL): Supports both PostgreSQL ($5/month single-node entry) and MySQL via Vitess, developer-friendly schema migrations with non-blocking branching, removed traditional free tier in 2024, Scaler Pro plan with pay-as-you-go consumption, PostgreSQL now generally available (GA) as of 2025. Turso (SQLite-based): Best for edge deployment, $4.99/month Developer plan with unlimited databases and 9GB storage, global edge replication across hundreds of locations, free tier: 100 databases + 5GB storage + 500M row reads/month, unique database-per-tenant multi-tenancy. Key differences: Neon = PostgreSQL + true scale-to-zero + instant branching; PlanetScale = PostgreSQL/MySQL + enterprise migrations + Vitess hyperscale; Turso = SQLite + edge proximity + unlimited databases. Choose: Neon for cost optimization + PostgreSQL branching, PlanetScale for schema migrations + dual database support, Turso for edge latency + multi-tenant isolation.

Copy-on-write (CoW) architecture: branching creates instant database copies (~1 second) with zero data duplication, no storage bloat. Development workflow: create database branches for development/testing/preview deployments, each branch = isolated environment with production data snapshot up to point of creation. Benefits: test schema changes without risk, parallel feature development with separate branches, instant rollback by switching branches, preview deployments per pull request, true dev/prod parity. Cost advantage: branches share unchanged data pages (CoW mechanism), only divergent changes consume storage (billed for unique data only), enables 10s or 100s of branches without 10x storage costs, branches scale-to-zero when inactive. Speed: branch creation in ~1 second regardless of database size (vs minutes/hours for traditional pg_dump/restore clones), branch-per-PR workflow becomes practical for continuous deployment. Real use case: developer creates branch from production, tests breaking schema migration, validates with team, merges to main, deletes branch—all in minutes with zero prod risk. vs Traditional: cloning production = slow + expensive + storage overhead; Neon branching = instant + cost-efficient + storage-optimal. Game changer for teams doing frequent schema changes, running ephemeral preview environments, or needing isolated testing with production data.

Turso (2025): SQLite-based edge database with global replication - unique architecture combining SQLite's embedded simplicity with distributed edge deployment via libSQL (open-source SQLite fork). Unique advantages: (1) Unlimited databases per plan: $4.99/month Developer plan supports unlimited databases with 500 monthly active database limit (competitors: PlanetScale $5/database single-node entry, Neon Launch $19/month for 100 projects). Enables database-per-tenant multi-tenancy at scale (1000 tenants = 1000 databases = $4.99/month if ≤500 active concurrently). (2) Edge replication: Automatic multi-region replication - create database, select regions (US-East, EU-West, Asia-Pacific), Turso replicates to all via Fly.io infrastructure (hundreds of edge locations), reads served from nearest edge (sub-10ms latency globally). Write propagation: eventual consistency (1-50ms sync lag between regions). (3) SQLite foundation: Embedded database (no separate server process), simple file-based storage, serverless-native (each request gets isolated SQLite instance), 100KB-10MB typical database size. (4) libSQL protocol: Wire-compatible with SQLite, adds edge features (replication, branching, point-in-time recovery, vector search). Edge replication architecture (2025): (1) Primary region: All writes go to primary (single source of truth), strong consistency within primary. (2) Edge replicas: Read-only replicas across hundreds of edge locations (Fly.io infrastructure), automatic syncing from primary (1-50ms lag), reads routed to nearest replica via GeoDNS. (3) Conflict resolution: Last-write-wins for eventual consistency, application handles conflicts if needed (rare with read-heavy workloads). Performance benchmarks (2025): - Read latency: 5-15ms from nearest edge replica (vs 100-300ms cross-region PostgreSQL queries). - Write latency: 20-50ms write to primary + replication to edges (typical), sub-40ms from five continents observed. - Throughput: SQLite handles 10K-50K reads/sec per replica (lightweight, no server overhead), 1K-5K writes/sec to primary. Cost model (2025): Free tier (Starter): 5GB storage, 100 databases, 100 monthly active databases, 500M row reads/month, 10M row writes/month - generous for prototypes/personal projects. Developer plan ($4.99/month): Unlimited databases (500 monthly active), 9GB storage + $0.75/GB overage, 2.5B row reads/month. Scaler plan ($24.92/month): Unlimited databases (2,500 monthly active), 24GB storage + $0.50/GB overage, 100B row reads. Use cases (2025): (1) Multi-tenant SaaS (database-per-tenant): Each customer gets isolated SQLite database - perfect data isolation, simple backups (export per customer), regulatory compliance (data residency per region). Example: 5,000 B2B customers × 100MB/customer = 500GB total, each tenant isolated database, $29/month Scaler plan. (2) Global edge applications: Next.js app deployed to Vercel Edge Functions (20+ regions), Turso database replicas co-located, <10ms database reads for users worldwide. Example: Blog platform - edge SSR reads posts from local Turso replica, writes (new posts) go to primary, propagate in seconds. (3) Edge compute + edge data: Cloudflare Workers + Turso - both run at edge, data co-located with compute, eliminates backend database latency. Example: E-commerce product catalog - Workers read from Turso edge (5ms), generate dynamic pages. (4) Read-heavy distributed workloads: 95% reads, 5% writes - perfect for Turso edge reads (fast), occasional writes (acceptable latency). Example: Analytics dashboard - queries read from edge replica (instant), aggregations precomputed and written to primary (hourly batch). Unique features (2025): (1) Database branching: Git-like branches for development - create branch from production, test schema changes, merge back. turso db create dev-branch --from production. (2) Point-in-time recovery: Restore database to any timestamp (up to 30 days retention). Useful for accidental deletes, compliance audits. (3) Embedded replicas: Download entire database to client (mobile app, desktop app), sync changes back - offline-first architecture. (4) SQL over HTTP: REST API for queries (POST /v2/execute with SQL), no persistent connections needed (serverless-friendly). Trade-offs vs PostgreSQL/MySQL (2025): SQLite limitations: (1) Concurrent writes: Single writer at a time (other writers wait), PostgreSQL supports 1000s of concurrent writes via MVCC. Turso: 1K writes/sec practical limit. (2) Feature set: No stored procedures, limited full-text search (FTS5 available but basic), no advanced indexing (GiST, GIN), no JSON operators like PostgreSQL. (3) Query planner: Simpler than PostgreSQL - complex joins (5+ tables) slower, no parallel query execution. When Turso better: (1) Embedded use case: Each user/tenant gets own database file - SQLite perfect for isolation. (2) Edge latency critical: <20ms database reads required globally - Turso edge replication only solution. (3) **Cost-sensitive at scale**: 1000 databases on PostgreSQL = $19K/month (Neon), Turso = $29/month. **When PostgreSQL better**: (1) **Complex transactions**: Multi-statement transactions with rollback, PostgreSQL ACID guarantees stronger. (2) **Advanced SQL features**: Window functions, CTEs, complex aggregations - PostgreSQL more powerful. (3) **Heavy concurrent writes**: >5K writes/sec sustained - PostgreSQL handles better. Migration path (2025): SQLite → Turso trivial (libSQL wire-compatible), PostgreSQL → Turso requires schema refactor (remove advanced features, redesign for eventual consistency). Production deployments (2025): Next.js on Vercel Edge + Turso (popular combo), Remix on Fly.io + Turso (co-located edge), Cloudflare Workers + Turso (fastest edge). 2025 best practices: (1) Database-per-tenant: Leverage unlimited databases for perfect isolation (security, compliance, performance). (2) Read replicas everywhere: Place replicas in all user regions (5-10ms local reads). (3) Batch writes: Combine writes into transactions (reduce round-trips to primary). (4) Monitor replication lag: CloudWatch/Datadog dashboards track edge sync delay (should stay <100ms). (5) Handle conflicts: Eventual consistency means rare conflicts - design idempotent operations, use timestamps for conflict resolution.

Turso is SQLite-based edge database with global replication combining SQLite's embedded simplicity with distributed edge deployment via libSQL (open-source SQLite fork). Unique advantages: (1) Unlimited databases per plan - $4.99/month Developer plan supports unlimited databases with 500 monthly active limit, enables database-per-tenant multi-tenancy (1000 tenants = $4.99/month if ≤500 active concurrently vs $19K/month with PostgreSQL). (2) Edge replication - automatic multi-region replication across hundreds of locations via Fly.io infrastructure, reads served from nearest edge (sub-10ms latency globally), write propagation with eventual consistency (1-50ms sync lag). (3) SQLite foundation - embedded database (no separate server process), serverless-native (each request gets isolated instance), 100KB-10MB typical database size. Edge replication architecture: (1) Primary region - all writes go to primary (single source of truth), strong consistency within primary. (2) Edge replicas - read-only replicas across hundreds of edge locations, automatic syncing from primary, reads routed to nearest replica via GeoDNS. (3) Conflict resolution - last-write-wins for eventual consistency, rare conflicts with read-heavy workloads. Performance benchmarks 2025: read latency 5-15ms from nearest edge (vs 100-300ms cross-region PostgreSQL), write latency 20-50ms to primary + replication, throughput 10K-50K reads/sec per replica (lightweight, no server overhead), 1K-5K writes/sec to primary. libSQL protocol: wire-compatible with SQLite, adds edge features (replication, branching, point-in-time recovery, vector search). Production deployments: Next.js on Vercel Edge + Turso (popular combo), Remix on Fly.io + Turso (co-located edge), Cloudflare Workers + Turso (fastest edge). Real-world example: Next.js blog deployed to 20 Vercel regions with Turso replicas co-located - edge SSR reads posts from local replica (5ms), writes (new posts) go to primary, propagate to all edges in seconds, global users get <10ms database reads.

Use Turso when: (1) Multi-tenant SaaS (database-per-tenant) - each customer gets isolated SQLite database, perfect data isolation, simple per-customer backups, regulatory compliance (data residency per region). Example: 5,000 B2B customers × 100MB = 500GB total, $29/month Scaler plan vs $19K/month PostgreSQL. (2) Global edge applications requiring <20ms database reads worldwide - Turso edge replication only solution delivering 5-15ms reads. Example: E-commerce catalog on Cloudflare Workers reads from co-located Turso edge (5ms). (3) Read-heavy workloads (95% reads, 5% writes) - perfect for Turso edge reads (fast), occasional writes (acceptable latency). Example: Analytics dashboard queries edge replica (instant), aggregations written to primary hourly. (4) Cost-sensitive at scale - 1000 databases on Turso = $29/month vs $19K/month Neon/PlanetScale. Use PostgreSQL/MySQL when: (1) Complex transactions - multi-statement transactions with rollback, PostgreSQL ACID guarantees stronger than SQLite single-writer model. (2) Advanced SQL features - window functions, CTEs, complex aggregations, stored procedures, full-text search (PostgreSQL more powerful). (3) Heavy concurrent writes >5K writes/sec sustained - PostgreSQL handles via MVCC, Turso limited to 1K writes/sec (single writer). SQLite limitations in Turso: (1) Concurrent writes - single writer at a time (others wait), PostgreSQL supports 1000s concurrent writes. (2) Feature set - no stored procedures, limited full-text search (FTS5 basic), no advanced indexing (GiST, GIN), no JSON operators. (3) Query planner - simpler than PostgreSQL, complex joins (5+ tables) slower, no parallel query execution. Trade-off matrix: Turso better for embedded use cases, edge latency critical, cost-sensitive at scale; PostgreSQL better for complex transactions, advanced SQL, heavy concurrent writes. Migration path: SQLite → Turso trivial (libSQL wire-compatible), PostgreSQL → Turso requires schema refactor (remove advanced features, redesign for eventual consistency). Decision framework: choose Turso if edge latency (<20ms) or database-per-tenant model is core requirement AND workload is read-heavy with simple transactions; choose PostgreSQL if need complex SQL, heavy writes, or strong ACID guarantees.

Cost model 2025: Free tier (Starter) - 5GB storage, 100 databases, 100 monthly active databases, 500M row reads/month, 10M row writes/month (generous for prototypes). Developer plan $4.99/month - unlimited databases (500 monthly active), 9GB storage + $0.75/GB overage, 2.5B row reads/month. Scaler plan $24.92/month - unlimited databases (2,500 monthly active), 24GB storage + $0.50/GB overage, 100B row reads. Key: unlimited databases enable database-per-tenant at fixed cost vs per-database pricing (PlanetScale $5/database). Unique features: (1) Database branching - Git-like branches for development, create branch from production turso db create dev-branch --from production, test schema changes, merge back. (2) Point-in-time recovery (PITR) - restore database to any timestamp up to 30 days retention, useful for accidental deletes, compliance audits. (3) Embedded replicas - download entire database to client (mobile/desktop app), sync changes back, offline-first architecture. (4) SQL over HTTP - REST API for queries (POST /v2/execute with SQL), no persistent connections needed (serverless-friendly). Production best practices: (1) Database-per-tenant strategy - leverage unlimited databases for perfect isolation (security, compliance, performance), each tenant gets own SQLite file. (2) Read replicas everywhere - place replicas in all user regions (5-10ms local reads), Turso automatically handles replication. (3) Batch writes - combine writes into transactions reducing round-trips to primary, SQLite single-writer model benefits from batching. (4) Monitor replication lag - CloudWatch/Datadog dashboards track edge sync delay (should stay <100ms), alert if exceeds 200ms. (5) Handle conflicts - eventual consistency means rare conflicts, design idempotent operations, use timestamps for conflict resolution. Real use cases: (1) SaaS platforms - database-per-customer isolation, simple backups, data residency compliance. (2) Edge applications - Next.js/Vercel + Turso for <10ms global reads. (3) Mobile apps - embedded replicas for offline-first, sync when online. Performance optimization: co-locate Turso replicas with compute (Vercel, Fly.io, Cloudflare), use SQL over HTTP for serverless functions (no connection pooling overhead), cache query results at CDN for static content. Cost optimization examples: 1000 customers on Turso Scaler ($29/month) vs 1000 Neon databases ($19K/month), massive savings for multi-tenant. Monitoring: track row reads/writes against plan limits, set alerts at 80% usage, upgrade before hitting caps to avoid throttling.

Neon (PostgreSQL): Best cost-effectiveness for variable workloads, usage-based pricing $5/month minimum (Launch plan), $0.106/CU-hour compute + $0.35/GB-month storage, automatic scale-to-zero after 5 min inactivity, copy-on-write branching (instantaneous, zero data duplication), free tier with 30 projects and 100 CU-hours per project. PlanetScale (PostgreSQL + MySQL): Supports both PostgreSQL ($5/month single-node entry) and MySQL via Vitess, developer-friendly schema migrations with non-blocking branching, removed traditional free tier in 2024, Scaler Pro plan with pay-as-you-go consumption, PostgreSQL now generally available (GA) as of 2025. Turso (SQLite-based): Best for edge deployment, $4.99/month Developer plan with unlimited databases and 9GB storage, global edge replication across hundreds of locations, free tier: 100 databases + 5GB storage + 500M row reads/month, unique database-per-tenant multi-tenancy. Key differences: Neon = PostgreSQL + true scale-to-zero + instant branching; PlanetScale = PostgreSQL/MySQL + enterprise migrations + Vitess hyperscale; Turso = SQLite + edge proximity + unlimited databases. Choose: Neon for cost optimization + PostgreSQL branching, PlanetScale for schema migrations + dual database support, Turso for edge latency + multi-tenant isolation.

Copy-on-write (CoW) architecture: branching creates instant database copies (~1 second) with zero data duplication, no storage bloat. Development workflow: create database branches for development/testing/preview deployments, each branch = isolated environment with production data snapshot up to point of creation. Benefits: test schema changes without risk, parallel feature development with separate branches, instant rollback by switching branches, preview deployments per pull request, true dev/prod parity. Cost advantage: branches share unchanged data pages (CoW mechanism), only divergent changes consume storage (billed for unique data only), enables 10s or 100s of branches without 10x storage costs, branches scale-to-zero when inactive. Speed: branch creation in ~1 second regardless of database size (vs minutes/hours for traditional pg_dump/restore clones), branch-per-PR workflow becomes practical for continuous deployment. Real use case: developer creates branch from production, tests breaking schema migration, validates with team, merges to main, deletes branch—all in minutes with zero prod risk. vs Traditional: cloning production = slow + expensive + storage overhead; Neon branching = instant + cost-efficient + storage-optimal. Game changer for teams doing frequent schema changes, running ephemeral preview environments, or needing isolated testing with production data.

Turso (2025): SQLite-based edge database with global replication - unique architecture combining SQLite's embedded simplicity with distributed edge deployment via libSQL (open-source SQLite fork). Unique advantages: (1) Unlimited databases per plan: $4.99/month Developer plan supports unlimited databases with 500 monthly active database limit (competitors: PlanetScale $5/database single-node entry, Neon Launch $19/month for 100 projects). Enables database-per-tenant multi-tenancy at scale (1000 tenants = 1000 databases = $4.99/month if ≤500 active concurrently). (2) Edge replication: Automatic multi-region replication - create database, select regions (US-East, EU-West, Asia-Pacific), Turso replicates to all via Fly.io infrastructure (hundreds of edge locations), reads served from nearest edge (sub-10ms latency globally). Write propagation: eventual consistency (1-50ms sync lag between regions). (3) SQLite foundation: Embedded database (no separate server process), simple file-based storage, serverless-native (each request gets isolated SQLite instance), 100KB-10MB typical database size. (4) libSQL protocol: Wire-compatible with SQLite, adds edge features (replication, branching, point-in-time recovery, vector search). Edge replication architecture (2025): (1) Primary region: All writes go to primary (single source of truth), strong consistency within primary. (2) Edge replicas: Read-only replicas across hundreds of edge locations (Fly.io infrastructure), automatic syncing from primary (1-50ms lag), reads routed to nearest replica via GeoDNS. (3) Conflict resolution: Last-write-wins for eventual consistency, application handles conflicts if needed (rare with read-heavy workloads). Performance benchmarks (2025): - Read latency: 5-15ms from nearest edge replica (vs 100-300ms cross-region PostgreSQL queries). - Write latency: 20-50ms write to primary + replication to edges (typical), sub-40ms from five continents observed. - Throughput: SQLite handles 10K-50K reads/sec per replica (lightweight, no server overhead), 1K-5K writes/sec to primary. Cost model (2025): Free tier (Starter): 5GB storage, 100 databases, 100 monthly active databases, 500M row reads/month, 10M row writes/month - generous for prototypes/personal projects. Developer plan ($4.99/month): Unlimited databases (500 monthly active), 9GB storage + $0.75/GB overage, 2.5B row reads/month. Scaler plan ($24.92/month): Unlimited databases (2,500 monthly active), 24GB storage + $0.50/GB overage, 100B row reads. Use cases (2025): (1) Multi-tenant SaaS (database-per-tenant): Each customer gets isolated SQLite database - perfect data isolation, simple backups (export per customer), regulatory compliance (data residency per region). Example: 5,000 B2B customers × 100MB/customer = 500GB total, each tenant isolated database, $29/month Scaler plan. (2) Global edge applications: Next.js app deployed to Vercel Edge Functions (20+ regions), Turso database replicas co-located, <10ms database reads for users worldwide. Example: Blog platform - edge SSR reads posts from local Turso replica, writes (new posts) go to primary, propagate in seconds. (3) Edge compute + edge data: Cloudflare Workers + Turso - both run at edge, data co-located with compute, eliminates backend database latency. Example: E-commerce product catalog - Workers read from Turso edge (5ms), generate dynamic pages. (4) Read-heavy distributed workloads: 95% reads, 5% writes - perfect for Turso edge reads (fast), occasional writes (acceptable latency). Example: Analytics dashboard - queries read from edge replica (instant), aggregations precomputed and written to primary (hourly batch). Unique features (2025): (1) Database branching: Git-like branches for development - create branch from production, test schema changes, merge back. turso db create dev-branch --from production. (2) Point-in-time recovery: Restore database to any timestamp (up to 30 days retention). Useful for accidental deletes, compliance audits. (3) Embedded replicas: Download entire database to client (mobile app, desktop app), sync changes back - offline-first architecture. (4) SQL over HTTP: REST API for queries (POST /v2/execute with SQL), no persistent connections needed (serverless-friendly). Trade-offs vs PostgreSQL/MySQL (2025): SQLite limitations: (1) Concurrent writes: Single writer at a time (other writers wait), PostgreSQL supports 1000s of concurrent writes via MVCC. Turso: 1K writes/sec practical limit. (2) Feature set: No stored procedures, limited full-text search (FTS5 available but basic), no advanced indexing (GiST, GIN), no JSON operators like PostgreSQL. (3) Query planner: Simpler than PostgreSQL - complex joins (5+ tables) slower, no parallel query execution. When Turso better: (1) Embedded use case: Each user/tenant gets own database file - SQLite perfect for isolation. (2) Edge latency critical: <20ms database reads required globally - Turso edge replication only solution. (3) **Cost-sensitive at scale**: 1000 databases on PostgreSQL = $19K/month (Neon), Turso = $29/month. **When PostgreSQL better**: (1) **Complex transactions**: Multi-statement transactions with rollback, PostgreSQL ACID guarantees stronger. (2) **Advanced SQL features**: Window functions, CTEs, complex aggregations - PostgreSQL more powerful. (3) **Heavy concurrent writes**: >5K writes/sec sustained - PostgreSQL handles better. Migration path (2025): SQLite → Turso trivial (libSQL wire-compatible), PostgreSQL → Turso requires schema refactor (remove advanced features, redesign for eventual consistency). Production deployments (2025): Next.js on Vercel Edge + Turso (popular combo), Remix on Fly.io + Turso (co-located edge), Cloudflare Workers + Turso (fastest edge). 2025 best practices: (1) Database-per-tenant: Leverage unlimited databases for perfect isolation (security, compliance, performance). (2) Read replicas everywhere: Place replicas in all user regions (5-10ms local reads). (3) Batch writes: Combine writes into transactions (reduce round-trips to primary). (4) Monitor replication lag: CloudWatch/Datadog dashboards track edge sync delay (should stay <100ms). (5) Handle conflicts: Eventual consistency means rare conflicts - design idempotent operations, use timestamps for conflict resolution.

Turso is SQLite-based edge database with global replication combining SQLite's embedded simplicity with distributed edge deployment via libSQL (open-source SQLite fork). Unique advantages: (1) Unlimited databases per plan - $4.99/month Developer plan supports unlimited databases with 500 monthly active limit, enables database-per-tenant multi-tenancy (1000 tenants = $4.99/month if ≤500 active concurrently vs $19K/month with PostgreSQL). (2) Edge replication - automatic multi-region replication across hundreds of locations via Fly.io infrastructure, reads served from nearest edge (sub-10ms latency globally), write propagation with eventual consistency (1-50ms sync lag). (3) SQLite foundation - embedded database (no separate server process), serverless-native (each request gets isolated instance), 100KB-10MB typical database size. Edge replication architecture: (1) Primary region - all writes go to primary (single source of truth), strong consistency within primary. (2) Edge replicas - read-only replicas across hundreds of edge locations, automatic syncing from primary, reads routed to nearest replica via GeoDNS. (3) Conflict resolution - last-write-wins for eventual consistency, rare conflicts with read-heavy workloads. Performance benchmarks 2025: read latency 5-15ms from nearest edge (vs 100-300ms cross-region PostgreSQL), write latency 20-50ms to primary + replication, throughput 10K-50K reads/sec per replica (lightweight, no server overhead), 1K-5K writes/sec to primary. libSQL protocol: wire-compatible with SQLite, adds edge features (replication, branching, point-in-time recovery, vector search). Production deployments: Next.js on Vercel Edge + Turso (popular combo), Remix on Fly.io + Turso (co-located edge), Cloudflare Workers + Turso (fastest edge). Real-world example: Next.js blog deployed to 20 Vercel regions with Turso replicas co-located - edge SSR reads posts from local replica (5ms), writes (new posts) go to primary, propagate to all edges in seconds, global users get <10ms database reads.

Use Turso when: (1) Multi-tenant SaaS (database-per-tenant) - each customer gets isolated SQLite database, perfect data isolation, simple per-customer backups, regulatory compliance (data residency per region). Example: 5,000 B2B customers × 100MB = 500GB total, $29/month Scaler plan vs $19K/month PostgreSQL. (2) Global edge applications requiring <20ms database reads worldwide - Turso edge replication only solution delivering 5-15ms reads. Example: E-commerce catalog on Cloudflare Workers reads from co-located Turso edge (5ms). (3) Read-heavy workloads (95% reads, 5% writes) - perfect for Turso edge reads (fast), occasional writes (acceptable latency). Example: Analytics dashboard queries edge replica (instant), aggregations written to primary hourly. (4) Cost-sensitive at scale - 1000 databases on Turso = $29/month vs $19K/month Neon/PlanetScale. Use PostgreSQL/MySQL when: (1) Complex transactions - multi-statement transactions with rollback, PostgreSQL ACID guarantees stronger than SQLite single-writer model. (2) Advanced SQL features - window functions, CTEs, complex aggregations, stored procedures, full-text search (PostgreSQL more powerful). (3) Heavy concurrent writes >5K writes/sec sustained - PostgreSQL handles via MVCC, Turso limited to 1K writes/sec (single writer). SQLite limitations in Turso: (1) Concurrent writes - single writer at a time (others wait), PostgreSQL supports 1000s concurrent writes. (2) Feature set - no stored procedures, limited full-text search (FTS5 basic), no advanced indexing (GiST, GIN), no JSON operators. (3) Query planner - simpler than PostgreSQL, complex joins (5+ tables) slower, no parallel query execution. Trade-off matrix: Turso better for embedded use cases, edge latency critical, cost-sensitive at scale; PostgreSQL better for complex transactions, advanced SQL, heavy concurrent writes. Migration path: SQLite → Turso trivial (libSQL wire-compatible), PostgreSQL → Turso requires schema refactor (remove advanced features, redesign for eventual consistency). Decision framework: choose Turso if edge latency (<20ms) or database-per-tenant model is core requirement AND workload is read-heavy with simple transactions; choose PostgreSQL if need complex SQL, heavy writes, or strong ACID guarantees.

Cost model 2025: Free tier (Starter) - 5GB storage, 100 databases, 100 monthly active databases, 500M row reads/month, 10M row writes/month (generous for prototypes). Developer plan $4.99/month - unlimited databases (500 monthly active), 9GB storage + $0.75/GB overage, 2.5B row reads/month. Scaler plan $24.92/month - unlimited databases (2,500 monthly active), 24GB storage + $0.50/GB overage, 100B row reads. Key: unlimited databases enable database-per-tenant at fixed cost vs per-database pricing (PlanetScale $5/database). Unique features: (1) Database branching - Git-like branches for development, create branch from production turso db create dev-branch --from production, test schema changes, merge back. (2) Point-in-time recovery (PITR) - restore database to any timestamp up to 30 days retention, useful for accidental deletes, compliance audits. (3) Embedded replicas - download entire database to client (mobile/desktop app), sync changes back, offline-first architecture. (4) SQL over HTTP - REST API for queries (POST /v2/execute with SQL), no persistent connections needed (serverless-friendly). Production best practices: (1) Database-per-tenant strategy - leverage unlimited databases for perfect isolation (security, compliance, performance), each tenant gets own SQLite file. (2) Read replicas everywhere - place replicas in all user regions (5-10ms local reads), Turso automatically handles replication. (3) Batch writes - combine writes into transactions reducing round-trips to primary, SQLite single-writer model benefits from batching. (4) Monitor replication lag - CloudWatch/Datadog dashboards track edge sync delay (should stay <100ms), alert if exceeds 200ms. (5) Handle conflicts - eventual consistency means rare conflicts, design idempotent operations, use timestamps for conflict resolution. Real use cases: (1) SaaS platforms - database-per-customer isolation, simple backups, data residency compliance. (2) Edge applications - Next.js/Vercel + Turso for <10ms global reads. (3) Mobile apps - embedded replicas for offline-first, sync when online. Performance optimization: co-locate Turso replicas with compute (Vercel, Fly.io, Cloudflare), use SQL over HTTP for serverless functions (no connection pooling overhead), cache query results at CDN for static content. Cost optimization examples: 1000 customers on Turso Scaler ($29/month) vs 1000 Neon databases ($19K/month), massive savings for multi-tenant. Monitoring: track row reads/writes against plan limits, set alerts at 80% usage, upgrade before hitting caps to avoid throttling.