From error-handling to structured concurrency

4 days ago

Error-handling in concurrent programs lacks a single stack for propagation, unlike single-threaded programs.
Two common approaches for unhandled errors in concurrent programs: terminate the thread and continue, or terminate the entire program.
Python's asyncio offers a third way by delivering exceptions to tasks waiting on the failed task, but risks swallowing exceptions if not handled.
Proposal: enforce waiting on every spawned task to handle exceptions, using context managers to ensure cleanup.
Challenges include deadlocks and resource leaks when tasks depend on each other or fail unexpectedly.
Solution: implement a cancellation mechanism to promptly respond to errors in any task, ensuring cleanup and preventing deadlocks.
Structured concurrency introduces a tree of tasks with nested lifetimes and automatic cancellation, improving error handling and resource management.
Structured concurrency frameworks like Python's asyncio TaskGroup, trio's nursery, and Go's errgroup package embody these principles.
Error-handling in concurrent programs is crucial not just for robustness but also for a smoother development experience, reducing frustration from deadlocks and swallowed exceptions.

Hasty Briefsbeta