Client Get operation stalls in getaddrinfo when network is down

[TheMostDiligent]

Hello!

We are seeing the following issue when the network is completely unavailable: the Get operations stalls in the getaddrinfo for about 30 seconds. We set the connection timeout and read timeout, but they don't have any effect.

Is it possible that this logic also respects the connection timeout?

[yhirose]

@TheMostDiligent thanks for the feedback. As far as I know, getaddrinfo is a blocking function and doesn't provide timeout option. So there is no way for me to deal with it... Do you know how other HTTP libraries handle this situation?

[TheMostDiligent]

I am not very profound in HTTP handling. However, I tried to research the subject a bit and apparently there is an alternative function: getaddrinfo_a that supports asynchronous wait mode (GAI_NOWAIT).
In this mode, the function exits immediately and an event is triggered when the operation completes. I think this will allow to wait for the event in the loop and abort the loop if timeout elapses. Or even better, if the user-provided callback returns false similar to the callback of the Get function.

[yhirose]

It looks like only Linux supports it... How could it be possible to do the same in Windows and MacOS?

[TheMostDiligent]

It looks like only Linux supports it

It does seem so. On Mac, there is DNSServiceGetAddrInfo that allows async wait. This article describes it in detail. Looks like it is quite a hassle to implement this...

On Windows, I actually did not see that problem.

[yhirose]

It seems like we may use GetAddrInfoEx which supports timeout parameter.
https://learn.microsoft.com/en-us/windows/win32/api/ws2tcpip/nf-ws2tcpip-getaddrinfoexw

In any case, supporting this feature on all three platforms and ensuring they work correctly will not likely be an easy task for me. So please don't expect me to implement it in the near future. A pull request is always welcome! :)

[yhirose]

This stackoverflow post is helpful to write thread timeout code.
https://stackoverflow.com/questions/40550730/how-to-implement-timeout-for-function-in-c

[TheMostDiligent]

@yhirose Thank you for fixing this!

[yhirose]

@TheMostDiligent I'm still testing it before releasing the next version. If the timeout parameter is set to 0, it uses the synchronous getaddrinfo. If the value is greater than 0, it attempts to use a native approach based on the user's operating system. If no native method is available, it falls back to a thread-based approach implemented in pure C++11.

Interestingly, there appears to be very little performance difference between the native and the C++11 thread-based approaches. Additionally, the native method introduces extra dependencies (Linux: -lanl, macOS: -framework CFNetwork).

Given this, I'm considering removing the native approach unless we find a significant performance advantage. This would simplify the code, make it fully cross-platform, and eliminate the need for users to link additional libraries.

[TheMostDiligent]

I would suggest sticking with native approaches. Using detached thread may result in issue under certain circumstances.