TurboRouter is an optional optimization layer on top of the Rust pipeline (which is always active). The Rust pipeline handles JSON streaming; TurboRouter adds query pre-registration to skip GraphQL parsing.
TurboRouter pre-registers GraphQL queries with their SQL templates, bypassing GraphQL parsing and validation for known queries. Combined with the Rust pipeline's JSON passthrough, this eliminates virtually all overhead between database and client.
| Stage | Basic Python GraphQL | TurboRouter + Table Views |
|---|---|---|
| Query Parsing | 0.5ms | ✅ 0ms (pre-registered) |
| Schema Validation | 0.3ms | ✅ 0ms (pre-validated) |
| Query Planning | 0.2ms | ✅ 0ms (pre-planned) |
| Hash Lookup (APQ) | - | 0.05ms |
| PostgreSQL Query | 15-30ms (JOINs) | 0.3-0.8ms (table view) |
| Deserialize → Python Objects | 2-8ms | ✅ 0ms (skipped) |
| Serialize Python Objects → JSON | 2-6ms | ✅ 0ms (skipped) |
| Rust Field Selection + Response | - | ✅ ~0.1ms |
| Total | 20-45ms | ✅ ~0.5-1ms |
Performance depends on query complexity and data size. The key speedup comes from table views (pre-computed JSONB) combined with TurboRouter's query pre-registration. Against optimized stacks with Redis caching, expect 2-3x improvement.
from fraiseql.fastapi import TurboQuery, TurboRegistry
registry = TurboRegistry()
user_query = TurboQuery(
graphql_query="""
query GetUser($id: ID!) {
user(id: $id) {
id
name
email
posts {
title
publishedAt
}
}
}
""",
sql_template="""
SELECT jsonb_build_object(
'user', jsonb_build_object(
'id', u.id,
'name', u.data->>'name',
'email', u.data->>'email',
'posts', COALESCE(
jsonb_agg(...)
)
)
) as result
FROM tb_user u
LEFT JOIN tb_post p ON p.fk_user = u.pk_user
WHERE u.id = %(id)s::uuid
GROUP BY u.id
""",
param_mapping={"id": "id"}
)
registry.register(user_query)# PostgreSQL returns JSONB:
{
"user": {
"id": 123,
"name": "Alice",
"email": "alice@example.com",
"posts": [...]
}
}
# TurboRouter: Stream it directly!
# ❌ No Python object creation
# ❌ No serialization overhead
# ✅ Zero-copy passthrough
# Application overhead: ~0.1ms ⚡Eliminate Python object instantiation and JSON serialization overhead
PostgreSQL JSONB flows directly to client without intermediate transformations
No temporary Python objects allocated—reduces GC pressure significantly
Handle 2-5x more requests per second with the same hardware
# Enable in production
app = create_fraiseql_app(
database_url="postgresql://...",
types=[User, Post],
production=True, # Enables TurboRouter
)
# Configure via environment
FRAISEQL_ENABLE_TURBO_ROUTER=true
FRAISEQL_TURBO_ROUTER_CACHE_SIZE=2000
# Or via config object
config = FraiseQLConfig(
enable_turbo_router=True,
turbo_router_cache_size=5000
)