InvalidNonce bitfinex2 invalid checksum about ccxt HOT 9 CLOSED

Hello @ssemeniuta,
Thanks for letting us know, we will take a look.

from ccxt.

@carlosmiei this issue might not be directly solvable , because of the frequency of incoming OB updates could be faster than the loop of await .watchOrderBook. so, the solution could be:
A) native ED subscriptions
or
B) handling exceptions in userland to ignore that message
or
C) saving incoming updates in cache (even if there are no clients in this.futures at the given moment) and once the await loopo returns back and re-subscribes into this.futures, then return the new data with respecting cached data

from ccxt.

@ttodua I'm not sure exactly how the frequency of incoming updates is related to the checksum validation.

from ccxt.

@ttodua I'm not sure exactly how the frequency of incoming updates is related to the checksum validation.

because the incoming hash (from exchange side) should match the hash of local orderbook, and if you miss some deltas to be integrated into the local orderbook, then you will end up in different orderbook data, therefore different hash

from ccxt.

@ttodua the orderbook is always updated regardless of the "user subscription", you can even call watch_order_book() only once and the orderbook will be updated in the background

from ccxt.

yes, you're right, what I wanted to interpret is that during frequent updates in bitfinex (if i correctly recall it was bitfinex) on same millisecond I have seen multiple incoming updates, thus when calculating the checksum of current orderbook, it might be still missing the first delta (because the whole handleMessage>handleOrderbook whole cycle can take more than 1 MS) thus, if on same MS there incomes two delta:

• A
• B

if B is followed by checksum validation, then A update might not be yet handled (thus missing during checksum).
I just theoretically assume this might be case, can't prove atm..

from ccxt.

@ttodua Thanks for the feedback. Do I understand correctly that your hypothesis is that book updates are too frequent to be fully processed by the script?

Atm I essentially do B) handling exceptions in userland to ignore that message from your suggestion - I let one script finish, wait for 2-3 minutes, then start the script again. There are periods when my app works for ~2-3 hours without this exception which is fine for now.

from ccxt.

@ssemeniuta I don't think that's the root cause for this issue, in the meantime while we debug it can you please try to disable the checksum validation by doing exchange.options['checksum'] = False ?

from ccxt.

I've modified the above code to

import ccxt.pro
from asyncio import run

print('CCXT Pro version', ccxt.pro.__version__)

params = {"prec": "P0"}
limit = 100

async def main():
    exchange = ccxt.pro.bitfinex2()
    await exchange.load_markets()
    exchange.verbose = False
    exchange.options['checksum'] = False
    symbol = 'AZERO/USDT'
    while True:
        try:
            orderbook = await exchange.watch_order_book(symbol, limit=limit, params=params)
            print(orderbook)
        except Exception as e:
            print(type(e).__name__, str(e))
            break

    await exchange.close()

run(main())

And the exception did not appear after ~24 hours of runtime. Thank you.

from ccxt.