[zeromq-dev] authentication failure and zmq_msg_send blocking

[Pieter Hintjens]

It's kind of the same problem as how to detect TCP connection failures
(DNS errors, network failures, etc.), and arguably the same answers.

On Tue, May 27, 2014 at 12:23 PM, Goswin von Brederlow
<goswin-v-b at web.de> wrote:
> On Mon, May 26, 2014 at 05:08:43PM +0200, Pieter Hintjens wrote:
>> Sorry for the slow reply. I assume you've passed this glitch, with the
>> latest master and have everything you need now?
>>
>> On Mon, May 19, 2014 at 3:19 PM, Goswin von Brederlow <goswin-v-b at web.de> wrote:
>> > Hi,
>> >
>> > I just tried building latest git and now 3 tests block:
>> >
>> > * test_security_null
>> > * test_security_plain
>> > * test_security_curve
>> >
>> > I only looked at the first so far but I believe all will be the same
>> > cause.
>> >
>> > test_security_null uses a DEALER socket and tests what happens when
>> > the authenication fails using expect_bounce_fail(). The expectation is
>> > that the send message is never recieved and gives a timeout since the
>> > client can't authenticate.
>> >
>> > But now the test blocks in zmq_send(), which doesn't have a timeout. I
>> > blieve this is a side effect of the recently merged work by hutonm
>> > concerning handing auth errors and reporting them to the client. My
>> > guess is this causes the DEALER socket to remain in the state where it
>> > has no available peers, which means zmq_send() blocks till there is
>> > one (which obviously never happens).
>> >
>> >
>> > This behaviour seems rather suboptimal to me. Wouldn't it make more
>> > sense for zmq_send() to return an error like EAUTH or ENOPEER when all
>> > peers of the socket have connected but got an error back? The
>> > connections will not be retried and wouldn't work the second time
>> > around (usualy) anyway. So zmq_send can never succeed and waiting
>> > seems pointless.
>> >
>> > Secondly how is the client supposed to detect authentication failures
>> > at all?
>> >
>> > MfG
>> >         Goswin
>
> The blockage is gone thanks to adding a timeout on send (my pull
> request and your minor fixup a bit later).
>
> But the problem remains how to detect authentication failures in the
> client.
>
> MfG
>         Goswin
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev