Fix Memory Leak in Flask

Memory Leaks in Flask

Flask's simplicity can mask memory leaks that grow with each request. Since Flask often runs in long-lived worker processes under Gunicorn, these leaks accumulate until the OOM killer intervenes.

Global State Accumulation

Storing data in module-level variables that grow per request:

# BAD — grows forever
request_log = []

@app.route('/api/data')
def get_data():
    request_log.append(request.args)  # Never cleared
    return jsonify(data)

Use Redis or a proper logging system instead.

Unclosed Database Connections

# BAD — connection never returned to pool
@app.route('/users')
def users():
    conn = engine.connect()
    result = conn.execute(text('SELECT * FROM users'))
    return jsonify([dict(r) for r in result])
    # conn.close() never called

# GOOD — use context manager
@app.route('/users')
def users():
    with engine.connect() as conn:
        result = conn.execute(text('SELECT * FROM users'))
        return jsonify([dict(r) for r in result])

Large Response Buffering

Streaming large files instead of loading them into memory:

# BAD — reads entire file into memory
@app.route('/download')
def download():
    data = open('large_file.csv').read()
    return data

# GOOD — stream it
@app.route('/download')
def download():
    def generate():
        with open('large_file.csv') as f:
            for line in f:
                yield line
    return Response(generate(), mimetype='text/csv')

Gunicorn Worker Recycling

As a safety net, restart workers after a number of requests:

# gunicorn.conf.py
max_requests = 1000
max_requests_jitter = 50

Bugsly tracks per-worker memory usage in Flask apps, alerting when a specific worker is growing disproportionately.