AHC may crash if an read idle timeout is set, but a running request fails before receiving http .end

[fabianfett]

Currently we only clear the idle read timer, if we receive an http .end.

async-http-client/Sources/AsyncHTTPClient/ConnectionPool/HTTP1.1/HTTP1ClientChannelHandler.swift

Line 468 in a0b0985

return .clearIdleReadTimeoutTimer

However if the request is cancelled or the connection fails, we may never receive such an event. For this reason we should ensure in

async-http-client/Sources/AsyncHTTPClient/ConnectionPool/HTTP1.1/HTTP1ClientChannelHandler.swift

Lines 34 to 35 in a0b0985

	private var request: HTTPExecutableRequest? {
	didSet {

with a willSet method that a running idleReadTimeoutTimer is cancelled. Currently we may run into a precondition if:

Currently we should run into a precondition, if the following happens:

1. On an HTTPClient that has idleTimeout enabled a request is sent
2. The remote server waits to receive the full request and closes the connection without sending a response
3. The HTTPClient and its ELG must be kept alive for at least the idleTimeout duration

We should run into the precondition here:

async-http-client/Sources/AsyncHTTPClient/ConnectionPool/HTTP1.1/HTTP1ConnectionStateMachine.swift

Line 311 in a0b0985

preconditionFailure("Invalid state: \(self.state)")

[artemredkin]

Does #455 fix this issue? (We've run into this crash with 1.6.x recently)

[Lukasa]

It should do, yes.

[dnadoba]

Fixed in #455 and released with 1.6.2