Feature Request Type

• Core functionality
• Alteration (enhancement/optimization) of existing feature(s)
• New behavior

Description

The current implementation of load_many on the dataloader uses asyncio.gather to run the batch of keys through the existing load implementation (its a single line):

def load_many(self, keys):
    return gather(*map(self.load, keys))

This means that if any of the individual load tasks raises an exception, the entire load_many call will fail. So for example, if key 1 returns a value but key 2 fails by raising, then this code:

results = await my_loader.load_many([1, 2])

raises an exception and the value for key 1 can't be used.

In some cases it would be useful to do:

results = await my_loader.load_many([1,2])

for result in results:
  if isinstance(result, Exception):
    # handle error
  else:
    # handle successful result

This would match the implementation of loadMany in JS dataloader project: https://github.com/graphql/dataloader#loadmanykeys

Implementation notes

The behaviour can be achieved with the return_exceptions argument to gather. For example:

def load_many(self, keys):
    return gather(*map(self.load, keys), return_exceptions=True)

https://docs.python.org/3/library/asyncio-task.html#asyncio.gather

Open questions

Adding return_exceptions in-place would change the behaviour of existing code. Another option would be to add a return_exceptions optional argument to the load_many method and allow clients to specify the behaviour (leaving the existing behaviour unchanged). I don't have a strong instinct either way.