MiddleDebuggingCommonNot answered yet

Why does this coroutine keep running after its scope was cancelled?

A repository call is wrapped in runCatching so that a network failure is logged instead of crashing the screen. But after the scope is cancelled the coroutine does not stop: it logs a bogus failure and calmly walks on to the next line.

Constraints: fetchUser is a suspend function doing a cancellable network call, real network failures must still be logged, and cancellation must actually stop the coroutine.

suspend fun load(id: String): User? =
    runCatching { fetchUser(id) }
        .onFailure { log.warn("load failed", it) }
        .getOrNull()

Find and fix the bug.

runCatching catches Throwable, so it also swallows the CancellationException that cancellation throws: the coroutine reads cancellation as an ordinary failure, logs it and carries on. Rethrow it, or catch only the exceptions you can act on.

✗Thinking runCatching catches only Exception and leaves CancellationException alone
✗Treating a caught CancellationException as a real failure worth logging and retrying
✗Assuming getOrNull() on a failed Result rethrows and thereby stops the coroutine

→Why does the same bug show up with a bare try / catch (e: Throwable)?
→What does calling coroutineContext.ensureActive() inside onFailure change here?

Contents

The bug

runCatching catches Throwable, not Exception. Coroutine cancellation arrives as a CancellationException, so it lands in the very same Result.failure:

suspend fun load(id: String): User? =
    runCatching { fetchUser(id) }            // catches CancellationException too
        .onFailure { log.warn("load failed", it) }  // logs cancellation as a failure
        .getOrNull()                          // the coroutine walks on regardless

Cooperative cancellation relies on that CancellationException flying all the way up to the builder. Swallowing it breaks the chain: the scope counts as cancelled while the coroutine keeps working.

The fix

Rethrow cancellation, log only genuine failures:

suspend fun load(id: String): User? =
    runCatching { fetchUser(id) }
        .onFailure { if (it is CancellationException) throw it }
        .onFailure { log.warn("load failed", it) }
        .getOrNull()

Cleaner still — do not catch Throwable at all, catch only what you can act on:

suspend fun load(id: String): User? =
    try {
        fetchUser(id)
    } catch (e: IOException) {              // a CancellationException never lands here
        log.warn("load failed", e)
        null
    }

Open topic in roadmap →

Contents

Why does this coroutine keep running after its scope was cancelled?

Constraints: fetchUser is a suspend function doing a cancellable network call, real network failures must still be logged, and cancellation must actually stop the coroutine.

suspend fun load(id: String): User? =
    runCatching { fetchUser(id) }
        .onFailure { log.warn("load failed", it) }
        .getOrNull()

Find and fix the bug.

The bug

runCatching catches Throwable, not Exception. Coroutine cancellation arrives as a CancellationException, so it lands in the very same Result.failure:

suspend fun load(id: String): User? = runCatching { fetchUser(id) } // catches CancellationException too .onFailure { log.warn("load failed", it) } // logs cancellation as a failure .getOrNull() // the coroutine walks on regardless

Cooperative cancellation relies on that CancellationException flying all the way up to the builder. Swallowing it breaks the chain: the scope counts as cancelled while the coroutine keeps working.

The fix

Rethrow cancellation, log only genuine failures:

suspend fun load(id: String): User? = runCatching { fetchUser(id) } .onFailure { if (it is CancellationException) throw it } .onFailure { log.warn("load failed", it) } .getOrNull()

Cleaner still — do not catch Throwable at all, catch only what you can act on:

suspend fun load(id: String): User? = try { fetchUser(id) } catch (e: IOException) { // a CancellationException never lands here log.warn("load failed", e) null }