binance: Retry keep alive. #2958
Conversation
JoeGruffins
force-pushed
the
binancewatchlistenkey
branch
3 times, most recently
from
f436428
to
7b64763
Compare
There was a problem hiding this comment.
Let's do something like this so that we can test.
Also, please look into the LIST_SUBSCRIPTIONS websocket message to see if we can leverage that in a periodic check to catch discrepancies.
|
Looking into querying the subscriptions. |
|
Sorry I have not tested this on testnet/mainnet yet, but I should be able to use on testnet correct? |
Maybe? I haven't found testing mm on testnet very useful. |
There was a problem hiding this comment.
Maybe a better technique would be if an orderbook has not received an update within a certain period, maybe 30 seconds, then resync the orderbook. Currently we will only know that an orderbook is stale every time checkSubs is run, so it could remain stale for 30 mins.
This would result in tons of unnecessary reloads on low-volume markets. |
There was a problem hiding this comment.
I've been using this diff to test on mainnet. I've turned off my internet connection for 20 mins, then turned it back on, and the orderbooks immediately resync without hitting the new code here. The code looks good otherwise and works fine with the simnet test. Would be nice to be able to reproduce the issue though.
JoeGruffins
force-pushed
the
binancewatchlistenkey
branch
3 times, most recently
from
56ed9ba
to
e3cf2a8
Compare
|
@martonp does the order book desync look ok? https://github.com/decred/dcrdex/compare/1a223db8b8ae83b3e4da70b6ddd1adbf421ef027..e3cf2a8358ff0ac4a6e6e61507439a2b1478f56b |
client/mm/libxc/binance.go
Outdated
| @@ -288,6 +291,11 @@ func (b *binanceOrderBook) Connect(ctx context.Context) (*sync.WaitGroup, error | |||
| if retry != nil { // don't hammer | |||
| continue | |||
| } | |||
| case <-b.disconnectedChan: | |||
There was a problem hiding this comment.
It looks like if for some reason the Websocket has not reconnected, but the snapshot request successfully completes, the orderbook will appear synced, but no messages will be coming through.
There was a problem hiding this comment.
I think the order book stuff doesn't happen over websocket? It's a separate secure http request? So we should stop the books from looking synced until the websocket looks up again?
There was a problem hiding this comment.
How about the last change https://github.com/decred/dcrdex/compare/7833026d42ae23acc577fd07548c97c2158e580f..adef014da6fd9c56e84197ce13a1a5e3a24d0147
So, not allowing sync again until the websocket tells us it is up. If the connect messages came out of order or we missed one that would be bad though, unsure if that can happen.
JoeGruffins
force-pushed
the
binancewatchlistenkey
branch
from
e3cf2a8
to
7833026
Compare
JoeGruffins
force-pushed
the
binancewatchlistenkey
branch
from
7833026
to
adef014
Compare
|
I don't think this test fail is due to my changes.... |
| // will not place new orders. | ||
| connected := cs == comms.Connected | ||
| bnc.booksMtx.RLock() | ||
| defer bnc.booksMtx.RLock() |
There was a problem hiding this comment.
| defer bnc.booksMtx.RLock() | |
| defer bnc.booksMtx.RUnlock() |
| select { | ||
| case reconnectC <- struct{}{}: | ||
| default: | ||
| } |
There was a problem hiding this comment.
What's the reason for this?
Untested. Attempting to solve a problem with stuck books.