withTimeoutOrNull and channel send/receive

[fvasco]

Our server is affected by sporadic errors, these errors are not well defined and not reproducible.

I wrote a bad test case (#1281) some time ago, sorry, a have to file a new issue.

In our previous experience the major candidate is withTimeoutOrNull + channel.send, the in that case the coroutine hangs indefinitely.

The prevision version below worked without any issue

val delay = launch { delay(42) }
whileSelect {
    channel.onReceive {
        consume(it) // no suspension
        true
    }
    delay.onJoin { false }
}

Yesterday we get a similar issue, the code is:

withTimeoutOrNull(42){
    val value = channel.receive()
    consume(value) // no suspension
}

We suspect that a single value was extracted but not consumed.

Thoughts?

Edit: consume is not a suspending function

[elizarov]

These particular code path are pretty well covered by functional and stress-tests and I wonder what might have gone wrong. Any chance for a reproducer?

[fvasco]

Thank you for response @elizarov,
this is wonderful response.

I tried to reproduce this issue without any success.

In this particular case the consume function is list.add(value), I detected in production that one item was lost or was not processed.
Unfortunately there is no log to understand what exactly went wrong.

I will reopen this issue when further updates will be available.

Thank you again.

[fvasco]

I answer myself.

Thoughts?

Bug or not, chunking a channel using withTimeoutOrNull is probably a bad style of code: it mixes cancellable and non-cancellable blocks, additionally it works using the non cancellability of non-suspending function as a trick.

Using a selector is the way to go, maybe it is slower and harder to write, moreover it requires to cancel delay job manually (#1065), but it works always as expected.

[elizarov]

Still it should work and pretty-well covered by tests....