Now web-handler task is cancelled on client disconnection (normal socket closing by peer or just connection dropping by unpluging a wire etc.)

This is pretty normal behavior but it confuses people who come from world of synchronous WSGI web frameworks like flask or django.

The problem is: if web-handler writes a data to DB the task might be cancelled without waiting for transmission completion. The following snippet might be stopped on every line:

async def handler(request):
    async with request.app['db'] as conn:
         await conn.execute('UPDATE ...')

To make execution atomic user should spawn a task:

async def writer(db):
    async with db as conn:
         await conn.execute('UPDATE ...')

async def handler(request):
    get_event_loop().create_task(writer(request.app['db']))

This approach is not reliable, new task is not awaited.

User need to use aiojobs:

async def handler(request):
    aiojobs.spawn(writer(request.app['db']))

Or maybe not existing yet:

@aiojobs.atomic
async def handler(request):
    async with request.app['db'] as conn:
         await conn.execute('UPDATE ...')

Sure, I could document it in http://aiohttp.readthedocs.io/en/stable/web.html but not sure is it very visible and obvious way to protect from cancellation. Ideas?