Fix MemoryError in FastAPI in Production

FastAPI Production Memory Errors

Production FastAPI services can slowly accumulate memory until the container is OOM-killed. Here's a systematic approach to finding and fixing the problem.

Profile Memory Usage

Add a debug endpoint (remove in production or protect it):

import tracemalloc
import os

tracemalloc.start()

@app.get("/debug/memory")
async def memory_stats():
    snapshot = tracemalloc.take_snapshot()
    top = snapshot.statistics('lineno')[:10]
    process_mb = os.popen('ps -o rss= -p %d' % os.getpid()).read().strip()
    return {
        "rss_mb": int(process_mb) / 1024,
        "top_allocations": [str(s) for s in top]
    }

Database Connection Pool

SQLAlchemy async sessions can leak if not properly scoped:

from sqlalchemy.ext.asyncio import create_async_engine, AsyncSession

engine = create_async_engine(
    DATABASE_URL,
    pool_size=10,
    max_overflow=5,
    pool_recycle=3600,  # Recycle connections hourly
    pool_pre_ping=True,
)

# Use dependency injection to ensure cleanup
async def get_db():
    async with AsyncSession(engine) as session:
        yield session

Container Configuration

resources:
  limits:
    memory: "512Mi"
  requests:
    memory: "256Mi"

Set Uvicorn's --limit-max-requests to recycle workers before they grow too large:

uvicorn main:app --workers 2 --limit-max-requests 5000 --limit-concurrency 50

Bugsly monitors process RSS over time and alerts when memory crosses configurable thresholds, giving you lead time before the OOM killer acts.