accept_no_activity_timeout seems obsolete but not removed. #7390
Description
Practical tests measuring the behaviour of accept_no_activity_timeout shows that is has no effect as documented.
The timeout interval in seconds before Traffic Server closes a connection that has no activity.
A test such as:
time telnet localhost 80
Show that the parameter that affects this timeout is:
proxy.config.http.default_inactivity_timeout
There was a suggestion that the parameter proxy.config.http.transaction_no_activity_timeout_in may affect it also.
However in practical tests I cannot achieve this.
It doesn't look as though it's possible to remain in the state where this parameter is in effect within the state machine for more than an instant. I cannot begin to imagine the failure type that would cause this to occur, and be required.
There is an intent/architectural question about whether a client that has never submitted data should have a different timeout than one that has previously been active. Personally I would use this if available, being more aggressive to this client type.
At the very least: This parameter and references to it should be removed, and descriptions of default_inactivity_timeout updated to include this case. However, I'm not certain this makes sense as the default timeout, being the least specific, is often longer. The use case for accept_no_activity_timeout is typically one of the most agressive.