Concurrency and Memory Safe System Software Development with Asynchronous Programming in Rust

E81 CSE 542S: Concurrency and Memory Safe System Software Development Asynchronous Programming Department of Computer Science & Engineering Chris Gill cdgill@wustl.edu 1

Asynchrony I Rust provides synchronous (std::) and asynchronous (async_std::) versions of many useful types and implementations Like net::TcpListener or net::TcpStream Block on calls to synchronous versions E.g., TcpListener::bind(addr); Vs. await asynchronous completion E.g., TcpListener::bind(addr).await; 2 CSE 542S Concurrency and Memory Safe System Software Development

Asynchrony II Asynchronous functions return futures Poll until function completes, returns result A.k.a. the Pi ata programming model Synchronous code can call asynchronous code conveniently in Rust using block_on I.e. block_on does the polling, returns result Can only await on futures from async functions/blocks within asynchronous code Doing so in synchronous code give an error 3 CSE 542S Concurrency and Memory Safe System Software Development

Asynchrony III Asynchronous code can nest and call/await multiple other async functions/blocks at once E.g., async fn run_server can call/await async_std::net::TcpListener::bind then spawn async blocks for accepted connections Each await lets asynchronous code suspend whenever a synch version would block Then, the next poll of its future resumes at the exact place where its execution was suspended Allows multiple asynchronous activities in parallel E.g., sending/receiving on the accepted connections 4 CSE 542S Concurrency and Memory Safe System Software Development

Asynchrony IV Spawning asynchronous code puts futures into a pool of threads that poll to ensure progress awaitcompletion in parallel, get each future s value vs. spawning synchronous threads and performing blocking joins on those threads handles For IO bound activities, spawn_local suffices Runs single core, can use thread-local storage For compute bound activities, use spawn instead Can run multi-core, must use task-local storage Also, all captured values must implement Send trait 5 CSE 542S Concurrency and Memory Safe System Software Development

Asynchrony V Performance of asynchronous code depends strongly on pipelining of data and control flows E.g., parallel execution of code on multiple cores E.g., parallel IO across NIC, filesystem, etc. Needs opportunities for different async blocks & functions to take turns using resources (safely) E.g., design async code so it calls yield_now reasonably frequently (whenever it makes sense) E.g., run code you don t have access to modify within a call to spawn_blocking which runs it in a separate thread and returns a future to await on E.g., Pin types implementing Deref or DerefMut 6 CSE 542S Concurrency and Memory Safe System Software Development

Studio 19 Download async-std crate on Linux Lab machines (will download/build locally, may take a while) Examine Rust s asynchronous programming features like async functions, await expressions, and block_on Practice using those features to develop and run simple server and client programs Studios 12 through 20 are due by 11:59pm on the night before Exam 2 Submit as soon as each is done so you get feedback and can resubmit any that may be marked incomplete 7 CSE 542S Concurrency and Memory Safe System Software Development