HTTP status code 408 causes clients to retry

[barancev]

Section 6.6 contains two error codes "script timeout"and "timeout" mapped to HTTP status code 408 Request Timeout.

It might be logically correct to use this error code, but we have a collision with HTTP semantics implemented in HTTP clients. To support keep-alive we allow retries in HTTP clients, and if a client sees code 408 it thinks that the server has gone and attempts to retry the request. As a results, we have executeScript or get/refresh duplication after timeout.

What can we do?

1. Stop using keep-alive, and hit the famous port exhaustion problem.
2. Stop using code 408 and replace it with a more neutral one that does not cause retries.

[andreastt]

Which HTTP code do you propose?

[barancev]

Unfortunately there is no HTTP code that would perfectly fit the purpose. We can use 400, as we do for the most other errors.

[barancev]

Or 500, because timeout is a server error, not a client one.

[andreastt]

I would prefer 500 over 400, since the latter gives the impression of a problem with the client’s request (e.g. malformed data or invalid arguments).

Can we get some input from @kereliuk, @burg, and @jimevans?

[andreastt]

Sorry, forgot @InstyleVII.

[burg]

I would prefer 500 over 400, per Andreas' argument.

[jimevans]

I don't know that I have a preference, but @andreastt makes a good argument. I'm perfectly happy using 500.

[andreastt]

For the interested, I also wrote a blog post about this change: https://sny.no/2018/08/webdriver-keepalive