We had a bug where if the payer is unusually well-connected compared to the payee, the payee is unable to accept a large payment since the payer will split it into lots of tiny payments, which would choke the max-concurrent-htlcs limit of the payee.
(node_factory, bitcoind)
| 4353 | |
| 4354 | @pytest.mark.slow_test |
| 4355 | def test_mpp_overload_payee(node_factory, bitcoind): |
| 4356 | """ |
| 4357 | We had a bug where if the payer is unusually well-connected compared |
| 4358 | to the payee, the payee is unable to accept a large payment since the |
| 4359 | payer will split it into lots of tiny payments, which would choke the |
| 4360 | max-concurrent-htlcs limit of the payee. |
| 4361 | """ |
| 4362 | # Default value as of this writing. |
| 4363 | # However, with anchor commitments we might be able to safely lift this |
| 4364 | # default limit in the future, so explicitly put this value here, since |
| 4365 | # that is what our test assumes. |
| 4366 | opts = {'max-concurrent-htlcs': 30} |
| 4367 | |
| 4368 | l1, l2, l3, l4, l5, l6 = node_factory.get_nodes(6, opts=opts) |
| 4369 | |
| 4370 | # Respect wumbo. |
| 4371 | # Using max-sized channels shows that the issue is not capacity |
| 4372 | # but rather max-concurrent-htlcs. |
| 4373 | # This is grade-school level. |
| 4374 | amt = 2**24 - 1 |
| 4375 | |
| 4376 | # Build the public network. |
| 4377 | # l1 is the very well-connected payer. |
| 4378 | # l2 is the poorly-connected payee. |
| 4379 | # l3->l6 are well-connected hop nodes. |
| 4380 | public_network = [l1.fundbalancedchannel(l3, amt), |
| 4381 | l1.fundbalancedchannel(l4, amt), |
| 4382 | l1.fundbalancedchannel(l5, amt), |
| 4383 | l1.fundbalancedchannel(l6, amt), |
| 4384 | l2.fundbalancedchannel(l6, amt), |
| 4385 | l3.fundbalancedchannel(l4, amt), |
| 4386 | l3.fundbalancedchannel(l5, amt), |
| 4387 | l3.fundbalancedchannel(l6, amt), |
| 4388 | l4.fundbalancedchannel(l5, amt), |
| 4389 | l5.fundbalancedchannel(l6, amt)] |
| 4390 | |
| 4391 | # Ensure l1 knows the entire public network. |
| 4392 | mine_funding_to_announce(bitcoind, [l1, l2, l3, l4, l5, l6]) |
| 4393 | for c in public_network: |
| 4394 | wait_for(lambda: len(l1.rpc.listchannels(c)['channels']) >= 2) |
| 4395 | |
| 4396 | # Now create a 400,000-sat invoice. |
| 4397 | # This assumes the MPP presplitter strongly prefers to |
| 4398 | # create lot sizes of 10,000 sats each. |
| 4399 | # This leads the presplitter to prefer to split into |
| 4400 | # around 40 HTLCs of 10,000 sats each, but since |
| 4401 | # max-concurrent-htlcs is set to 30, l2 would be unable |
| 4402 | # to receive. |
| 4403 | inv = l2.rpc.invoice(Millisatoshi(400000 * 1000), 'i', 'i')['bolt11'] |
| 4404 | |
| 4405 | # pay. |
| 4406 | l1.rpc.pay(inv) |
| 4407 | |
| 4408 | |
| 4409 | def test_offer(node_factory, bitcoind): |
nothing calls this directly
no test coverage detected