fix(turso): initialize runtime sqlite pragmas for local db

[dustintownsend]

Summary

• configure runtime local SQLite connections in Turso world with PRAGMA journal_mode = WAL and PRAGMA busy_timeout = 5000
• run pragma initialization once per cached client key to avoid duplicate initialization work
• add a new unreleased changeset for @workflow-worlds/turso only

Why

busy_timeout is connection-scoped. The setup CLI already applies pragmas on its own connection, but runtime world connections also need this configuration. Without it, upstream Turso e2e can hit transient SQLITE_BUSY: database is locked failures that cascade into timeouts.

Validation

• pnpm --filter @workflow-worlds/turso test

[github-actions]

Benchmark Comparison

Cross-comparison of workflow performance across World implementations.

📈 Comparing against baseline from main branch. Green 🟢 = faster, Red 🔺 = slower.

workflow with no steps

World	Workflow Time	Wall Time	Overhead	vs Fastest
🥇 💾 Starter (in-memory)	0.015s (+1.4%)	1.006s (~)	0.992s	1.00x
🔴 Redis	0.020s (+1.0%)	1.009s (~)	0.989s	1.41x
💻 Local	0.033s (+19.4% 🔺)	1.007s (~)	0.974s	2.25x
🗄️ Turso	0.088s (+139.9% 🔺)	1.011s (~)	0.924s	6.06x
🍃 MongoDB	0.091s (+97.0% 🔺)	1.010s (~)	0.919s	6.28x

workflow with 1 step

World	Workflow Time	Wall Time	Overhead	vs Fastest
🥇 💾 Starter (in-memory)	0.043s (+4.1%)	1.005s (~)	0.961s	1.00x
🔴 Redis	0.056s (+0.5%)	1.008s (~)	0.952s	1.29x
💻 Local	0.080s (-17.9% 🟢)	1.006s (~)	0.926s	1.83x
🍃 MongoDB	0.266s (-3.9%)	1.009s (~)	0.743s	6.12x
🗄️ Turso	0.286s (+18.1% 🔺)	1.010s (-1.2%)	0.724s	6.58x

workflow with 10 sequential steps

World	Workflow Time	Wall Time	Overhead	vs Fastest
🥇 💾 Starter (in-memory)	0.338s (+0.8%)	1.005s (~)	0.666s	1.00x
🔴 Redis	0.425s (~)	1.007s (~)	0.582s	1.26x
💻 Local	0.639s (-1.6%)	1.006s (~)	0.367s	1.89x
🗄️ Turso	2.112s (~)	3.012s (~)	0.900s	6.24x
🍃 MongoDB	2.143s (+0.6%)	3.011s (~)	0.868s	6.33x

workflow with 10 parallel steps

World	Workflow Time	Wall Time	Overhead	vs Fastest
🥇 💾 Starter (in-memory)	0.139s (+7.1% 🔺)	1.004s (~)	0.865s	1.00x
🔴 Redis	0.170s (+4.4%)	1.007s (~)	0.838s	1.22x
💻 Local	0.275s (-0.8%)	1.006s (~)	0.730s	1.98x
🍃 MongoDB	1.539s (-1.5%)	2.008s (~)	0.469s	11.07x
🗄️ Turso	1.590s (+0.9%)	2.010s (~)	0.420s	11.44x

---

Stream Benchmarks

Stream benchmarks include Time to First Byte (TTFB) metrics.

workflow with stream

World	Workflow Time	TTFB	Wall Time	Overhead	vs Fastest
🥇 💾 Starter (in-memory)	0.071s (-0.7%)	1.002s (~)	1.005s (~)	0.934s	1.00x
🔴 Redis	0.091s (+2.1%)	1.002s (~)	1.007s (~)	0.915s	1.29x
💻 Local	0.138s (-1.4%)	0.992s (~)	1.009s (~)	0.871s	1.94x
🍃 MongoDB	0.467s (-5.9% 🟢)	0.962s (+2.8%)	1.008s (~)	0.541s	6.61x
🗄️ Turso	0.486s (+9.4% 🔺)	0.947s (-4.0%)	1.009s (~)	0.523s	6.88x

---

Summary: Average Performance by World

World	Avg Workflow Time	Benchmarks
💾 Starter (in-memory)	0.121s	5
🔴 Redis	0.153s	5
💻 Local	0.233s	5
🍃 MongoDB	0.901s	5
🗄️ Turso	0.913s	5

Column Definitions

• Workflow Time: Runtime reported by workflow (completedAt - createdAt) - primary metric
• TTFB: Time to First Byte - time from workflow start until first stream byte received (stream benchmarks only)
• Wall Time: Total testbench time (trigger workflow + poll for result)
• Overhead: Testbench overhead (Wall Time - Workflow Time)
• vs Fastest: How much slower compared to the fastest World for this benchmark

Worlds:

• 💻 Local: Built-in workflow world
• 🐘 Postgres: PostgreSQL database world
• ▲ Vercel: Vercel production world
• 💾 Starter: In-memory reference implementation
• 🍃 MongoDB: MongoDB database world
• 🔴 Redis: Redis/BullMQ backend

[github-actions]

❌ E2E Test Results

World	Passed	Failed	Skipped	Total
MongoDB	44	1	3	48	❌
Redis	45	0	3	48	✅
Turso	45	0	3	48	✅
Total	134	1	9	144

Failed Tests

MongoDB — 1 failed

• e2e webhookWorkflow

  AssertionError: expected '' to be 'Hello from webhook!' // Object.is equality

View workflow run