[#569] fix stored connections to match desired connection timeout #570
Conversation
d-w-moore
force-pushed
the
569.fix_stored_connection_timeouts
branch
from
79ba79d
to
f501c20
Compare
Delete the check against connection_timeout parmeter with value of None. A `None` should be allowed as this is the Pythonic indicator for altogether disabling timeouts on a socket connection.
d-w-moore
force-pushed
the
569.fix_stored_connection_timeouts
branch
from
f501c20
to
ee0e215
Compare
| @@ -350,8 +350,7 @@ def connection_timeout(self): | |||
| @connection_timeout.setter | |||
| def connection_timeout(self, seconds): | |||
| self._cached_connection_timeout = seconds | |||
| if seconds is not None: | |||
There was a problem hiding this comment.
Many systems also use 0 to indicate a non-value (or forever)… consider checking for 0 or None?
There was a problem hiding this comment.
That is a good point as evidently 0 means set nonblocking, and I believe we want to disallow that.
None , however, in my opinion should be permissible as it is (for all intents) equivalent to a very large number.
There was a problem hiding this comment.
In other words session.connection_timeout = None is clear and correct in intent, if you take it to mean "eliminate timeouts totally for the given session."
There was a problem hiding this comment.
Agreed that None conveys the intent.
I was suggesting that if this is a user-facing setting, that setting it to either 0 or None would be equivalent and correct. If this is not user-facing, then... carry on.
There was a problem hiding this comment.
The docs now encourage the use of:
session_object.connection_timeout = <some_number>
as a way to dictate how long the client socket should wait until declaring a server peer as down and giving up, as it were.
So, yes, it is user-facing....
But 0 and None are not equivalent here, actually. 0 would invalidate the socket for further use by the PRC, by turning on non-blocking behavior (ie null-length values could now be read from the socket when it has not yet received any bytes of the expected server response) whereas None declares we never want to declare the server peer down. i.e. "turn off timeouts completely"
There was a problem hiding this comment.
If Windows is a different case, though, I should re-evaluate
There was a problem hiding this comment.
No worries - if None is the only correct answer - please consider leaving a comment somewhere user-facing... that will let them know that 0 behaves differently and why.
There was a problem hiding this comment.
Sounds good. I'll update the docs to caution against setting 0 (which doesn't yet cause an immediate error , but should) or even too low a value, as the latter would have an effect down the line of raising a NetworkError the first time the server "responds too slowly"... which (as long as the server is alive and still working) would be technically incorrect practice on the application writer's part.
Needs a test or two.
(Tests should include evaluating whether the new code works for SSL sockets.)