multiple concurrent connext transfers causes weird issues

[raladev]

Steps to reproduce

0. U have enough balances for both sides
1. Place ten buy 0.1 LTC/ETH 1 orders
2. Fill them as a maker sell 1 LTC/ETH mkt

Actual behaviour

Swap timeout after third swap + xud of maker is crashed after that + connext client of maker stops print new logs Screenshot from 2020-06-26 15-41-47

Expected behaviour

1. All orders should be filled
2. No xud crash because of misbehavior of connext client

Logs

part_fall_lndltc_taker.log part_fall_connext_taker.log part_fall_xud_taker.log

part_fall_lndltc_maker.log part_fall_connext_maker.log part_fall_xud_maker.log

[raladev]

@erkarl there is also stranges in connext balances of both nodes.

1.

Maker channel balances before swap: ETH 3.5 LTC 2.1996338

Taker channel balances before swap: ETH 3.65 LTC 2.8

Screen: balances+swaps

2. 10 swaps were started and 3 of them were successful. We can see it in tradehistory, xucli and lndltc-lncli listchannels outputs (it has only 7 pending htlcs).

Screens: lndltc-lncli_listchannels tradehistory_of_both_sides

3. So, lets check balance changes after swaps.

|LTC

Side	Before Swap	After Swap	Balance Change
Maker	2.199638	2.499036	2.499036-2.199638=0.299398
Taker	2.8	1.8	1.8-2.8=-1

ETH

Side	Before Swap	After Swap	Balance Change
Maker	3.5	2.6	2.6-3.5=-0.9
Taker	3.65	4.05	4.05-3.65=0.4

LTC is fine for both users in general (0.3 deposited, 0.7 in pending htlcs). Only 0.3-0.299398=0.000602 disappeared with no reason.

But ETH is totally bad.

1. only 0.9 ETH were blocked instead of 1 ETH
2. Taker got 0.4 ETH, when only 3 swaps were succesful.

[erkarl]

These logs are super interesting. Thanks for the additional info. I'm trying to figure out what happened.

[erkarl]

So, what happened here was:

• one side created bunch of orders
• other side created 1 order that matched them all in quantity

Result: => all the concurrent transfers see that there is collateral in the channel and insert into the queue and the first one to complete will take the collateral and others will fail since there is no more collateral

[erkarl]

This issue can be separated into 2:

• my above comment about what happened (TBD on how to handle these scenarios)
• xud should not crash after a failed swap => moved to https://github.com/ExchangeUnion/xud/issues/1701

[erkarl]

Lowering the priority of this since it's not causing xud to crash any more and we need to discuss on how to solve this.

[kilrau]

Just to clarify the analysis in https://github.com/ExchangeUnion/xud/issues/1691#issuecomment-650287939: this only happens for multiple concurrent connext transfers of the same client.