Can you identify any idioms or approaches that seem normal or natural in other languages, but which are problematic in Rust?
A similar situation occurred in the asynchronous communication with gopher servers. I use a thread to fetch a resource from another server while keeping the ui responsive. Passing the result from the thread back did not work out in the beginning.
I don’t think the idioms are so different in Rust, but when passing data between components, threads and the like, you have to be very specific in your code, using Arcs, RwLocks, Mutexes, etc.
On the bright sight, when the code compiles, it usually worked as intended. Using C or C++, there would be more race conditions, null pointer exceptions and more.
Async/await syntax is poorly documented across the ecosystem but has landed now and is useful, without it codebases will look outdated soon enough, hyper itself has some of the best docs
Lazystatic is popular way to use globals like that yet don't really bother with it anymore. The paradigms are a bit different for a language, you are discouraged from such things for a reason, there's certainly ways around using globals.
You don't need extern crate with 2018 edition and macros are automatically imported
The symbol should be allocated once on the book. The side is allocated twice in each set of vecs. The price is defined in the vec itself. The queue should be a vec of IDs and open size. The real struct should only be constructed lazily as absolutely needed.
This is a cool project btw - I'm looking for my first chance to really sink my teeth into rust so found it interesting. A few non-rust observations (you probably know all this but kept your orderbook simple which is totally cool but just in case):
An actual order book has a "tick ladder" which specifies which price points are valid for a given ticker on a given venue. This means you actually know all the possible prices ahead of time. If you ever wanted to productionize this you could probably take advantage of that and make a custom data structure that used the fact that you had all the keys preset and did the insert and lookup in O(1).
Secondly in real life the matching logic is quite a bit more complex because of order flags and venue logic. If you're just covering US venues you don't need to worry about venues because of Reg NMS but anywhere else you'll want the order books to actually by venue and symbol not just symbol.
Order flags interfere with matching in a number of ways. Firstly a normal limit order will partially match up to the limit and any remaining quantity will sit on the book waiting to be matched. You can flag the order as "fill or kill" to say fill the whole order or nothing and you can flag it as "immediate or cancel" to say don't put any remaining quantity passively on the book. There are many other order flags but those are probably the most important ones for matching.
Secondly on many venues orders can have an iceberg order type which means that only the "tip" quantity gets shown on the book but they will actually match up to a larger quantity at that price.
Finally you can add a "minimum executed quantity" on any passive orders to prevent pinging (information leakage). This will only match with incoming orders bigger than the MXQ, so messes with the normal price, time ordering. Orders behind that order could match first potentially. MXQ is more important for dark venues or order types so maybe you don't need to consider it.
[1] https://doc.rust-lang.org/std/collections/struct.BTreeMap.ht...
[2] https://doc.rust-lang.org/std/collections/struct.BinaryHeap....
Can you elaborate a little on the "pinging" process to extract information? Sounds interesting
Anyway the way MXQ and pinging works is imagine you have an orderbook where passive orders are not visible to other participants (ie the venue is dark or you have a grey venue where some orders can be flagged as dark). It's possible fro participants to pay to learn what orders are in the book by making some very small aggressive orders and seeing where they match. In the world of equities this would be buy 1 share agressively, sell 1 share agressively with a limit in each case. This is called "pinging" because it's like ICMP ping discovering whether hosts are alive. The concept in crypto (if it applies) would be the smallest quantity the exchange allows you to buy or sell.
Assuming the venue supports it, you can prevent your dark orders matching these pings by setting an MXQ. If your MXQ is 100 say, they would need to buy 100 shares for your order to match. You would set the size to a point where you don't mind giving up the information about where your order level is because the match is meaningful enough.
[1] https://github.com/binance-exchange/binance-official-api-doc...
Edit: maybe non-penny-pilot options trading > $3 are a better example of a product with a large tick size.