the root issue is i may not have enough available memory to read from the
socket.
I'll do the simple thing and just avoid muting handshaking sockets, for now.
On Wed, Nov 2, 2016 at 6:59 PM, Harsha Chintalapani <ka...@harsha.io> wrote:
> HI Radai,
> One main reason is to keep the handshake details away from
> the application layer. i.e Kafka network layer which is sending Kafka
> protocols doesn't need to worry about the handshake details, all it needs
> is a validation that the connection is completed and it can start sending
> Kafka protocols over the wire. So when a client tries to connect to a
> broker's SSL port, it goes through the handshake, if we mute the channel,
> Kafka network need to decide when to unmute it that means leaking some of
> the SSL connection details in Kafka Selector code. Given that we are
> supporting multiple Secure channels and each has its handshake mechanism we
> kept the selector code the same irrespective of which channel/port/security
> its trying to use. The details will be handled by the TransportLayer its
> job is to finish the handshake and return the ready() to be true when its
> ok for the client to start sending requests.
> As Joel said, it's possible to pause/resume the handshake but
> not sure why its needed; you can treat that as a black box and start
> sending your requests once the channel.ready(). I haven't gone through
> KIP-72 proposal so I might be missing something here.
>
> Thanks,
> Harsha
>
> On Wed, Nov 2, 2016 at 5:01 PM Joel Koshy <jjkosh...@gmail.com> wrote:
>
> > Sriharsha can validate this, but I think the reason is that if we allow
> > muting/unmuting at will (via those public APIs) that can completely mess
> up
> > the handshake itself. It should be possible to pause/resume the handshake
> > if that's what you'r elooking for but I'm not sure it is worth it for the
> > purposes of KIP-72 given the small volumes of reads/writes involved in
> > handshaking.
> >
> > On Wed, Nov 2, 2016 at 4:24 PM, radai <radai.rosenbl...@gmail.com>
> wrote:
> >
> > > Hi,
> > >
> > > as part of testing my code for KIP-72 (broker memory control), i ran
> into
> > > the following code snippet in SslTransportLayer:
> > >
> > > public void removeInterestOps(int ops) {
> > > if (!key.isValid())
> > > throw new CancelledKeyException();
> > > else if (!handshakeComplete)
> > > throw new IllegalStateException("handshake is not
> > completed");
> > >
> > > key.interestOps(key.interestOps() & ~ops);
> > > }
> > >
> > > why cant an ssl socket be muted before handshake is complete?
> > >
> >
>
