AI summary by TwinMind
How I Deploy ALL My Apps - My 2023 Stack
Theo - t3․gg
7m 26s
Introduction & Philosophy
- Third generation infrastructure tooling needed after AWS (launched 2006)
- Focus on developer experience over traditional server management
- Move from thinking about servers to thinking about functions (inputs/outputs)
- Created T3 deployment partner program - sponsorship from companies actually used
Core Serverless Platform - Vercel
- Core of almost every application shipped
- Eliminates need for Kubernetes, Terraform (abstraction leaks for most apps)
- Cannot remember last deployment not running on Vercel
- Simple deployment: sign in with GitHub, click repo, auto-configures everything
- Supports AWS Lambda serverless functions
- Supports Cloudflare edge deployment
- Handles CDN automatically
- Covers all code-running needs that aren't traditional servers
Database Solution - PlanetScale
- Stores "pretty much everything" now
- Ended chronic database hopping (Worker KV → Heroku Postgres → AWS Postgres → PlanetScale)
- MySQL-based (initially seemed scary but worth it)
- Two key strengths:
- Scale: never worry about reads/writes again
- Integrations: git-like workflow for migrations
- Each developer gets own branch for schema changes
- Deploy requests similar to pull requests for production merges
- NPM package for HTTP endpoint connection (no SQL connection needed)
- Enables edge deployment with instantaneous responses
- Eliminates connection formation wait times
Logging & Monitoring - Axiom
- Central logging tool for catching errors
- Two-click integration with Vercel
- All logs go to Axiom dashboard automatically
- Superior search functionality for finding errors
- Tracks what/where/when/why of issues
- At Ping: targets 7-minute response time (bug report → production fix)
- Wouldn't be possible without Axiom
- Deep Next.js understanding
- Provides Next.js package for client-side logging
- Bypasses ad blockers (u-block, etc.)
- Essential for serverless deployments
Additional Services - Upstash
- Handles crons, caches, events, message queues, rate limiting
- Started with making Redis easier to use
- Redis: in-memory database, fastest data retrieval, great for caching
- Traditional Redis: difficult to deploy, hard to use with TypeScript
- Upstash solutions:
- Rate limiting by IP address
- Dynamic cron jobs triggered by user actions
- Caching layer
- Event queues
- Provides helpful NPM packages
- Enables serverless-compatible infrastructure
- Prevents need to fall back to AWS
- Most painful 2023 decisions: chose AWS over Upstash for "professional" reasons
Authentication - Clerk (Late Addition)
- Last-minute addition to T3 deploy partners
- Previously recommended rolling own auth
- Problems with preview deploys, React Native support led to Clerk adoption
- Excellent mobile support, especially for create-T3-turbo
- Outstanding Next.js + Vercel experience
- Hard to recommend other solutions for this stack
- Auth provider of choice
Additional Tools
- GitHub (essential, irreplaceable)
- Other companies used but not as essential as core four
- Most could be swapped out except GitHub
