Just FWIW - we cut the 1.14 branch of native *prior to* merging the boost::asio change, so we would have some mileage on the change before releasing. So, all other things being equal, this will be in native 1.15, not 1.14.
Thanks, Blake -----Original Message----- From: Jacob Barrett <jabarr...@vmware.com> Sent: Wednesday, February 24, 2021 7:38 AM To: dev@geode.apache.org Subject: Re: Question about closing of all connections towards an endpoint in C++ native client Oh good! 1.14 is on its way, so problem solved! ;) > On Feb 24, 2021, at 7:31 AM, Alberto Gomez <alberto.go...@est.tech> wrote: > > Thanks Jake. I totally agree with you. > > Interestingly, that logic has been recently removed from the C++ client when > we switched from ACE_SOCK to boost::asio so what I said in my previous e-mail > pertained to the C++ client 1.13 version and older. > > Alberto > ________________________________ > From: Jacob Barrett <jabarr...@vmware.com> > Sent: Wednesday, February 24, 2021 4:24 PM > To: dev@geode.apache.org <dev@geode.apache.org> > Subject: Re: Question about closing of all connections towards an endpoint in > C++ native client > > The Java client does the same thing under certain conditions. Neither of the > clients should do this though. I think this model is way too overaggressive. > I think we should remove that logic entirely. If we think we want something > that proactively checks the other connections to that server we could have a > background thread go through and send a ping request on one the next in the > queue. If it doesn’t respond then terminate that connection. Continue until a > pong response is received. > > -Jake > >> On Feb 24, 2021, at 4:36 AM, Alberto Gomez <alberto.go...@est.tech> wrote: >> >> Hi, >> >> Running some tests with the C++ native client and looking at the code, I >> have observed that when an error in a connection towards an endpoint >> (timeout, IO error) is detected, not only the faulty connection is closed >> but the endpoint is set to "not connected" status which eventually provokes >> that all other open connections towards that endpoint are closed when used. >> >> I have not seen that behavior in the Java client, i.e., the Java client, >> when it detects an error in a connection towards an endpoint, it closes that >> connection but does not act on other connections towards that endpoint. >> >> Are my observations correct? >> >> If so, shouldn't the C++ native client be aligned with the Java client? >> >> Thanks, >> >> Alberto G. >
