I hope the colored text will work.
cf. below

On Fri, Apr 7, 2017 at 4:40 PM, Alan Conway <[email protected]> wrote:

> I jumped into the middle of this, sorry for irrelevant answers.
>
> Short story: I agree with you.
> Long story:
>
> On Thu, 2017-04-06 at 23:38 +0200, Olivier Mallassi wrote:
>
> > I ran the following tests:
> >
> > Access the broker directly (globalAddressDomains=/domain/subdomain +
> > queue1)
> > From a JMS client, the jndi queue name must contain the "/" .
> > (e.g. jndi name --> queue.myQueueLookup = /domain/subdomain/queue1)
> >
> >
> [snip]
> >
> >
> > From python, the "//" is needed.
> >
> > python ./simpleSend.py -a localhost:5672//domain/subdomain/queue1
>
> That's what I thought, good!
>
> The dispatch router has been configured with address and autoLink to
> /domain/subdomain/queue1. The AMQP Adress is normalized, hence
> > removing the first "/"
> [snip]
>
> I missed this part - you're talking about dispatch modifying addresses
> from the configuration file, not from a client.
>

Indeed. there is no client connected here.
Yet, I tried this because of the python example above.


>
> > Nos the dispatch router is configured with address and autoLink to
> > //domain/subdomain/queue1 (double "/")
> [snip]
> >
> > the //domain is removed.
> >
>
> The original AMQP addressing draft (from when dispatch started) defined
>  special meaning for leading '//' - to indicate connection information
> like this:
>
> //host:port/addr
>
> I think that is why dispatch is removing //domain.
>

This is also my guess.
I tried looking at the code.
When the link is activated (route_control.c#161) in the
method qdr_auto_link_activate_CT,
the following line is executed (L 182) qdr_terminus_set_address(term,
&key[2]);. In that case, the key[2] contains M1subdomain/queue1. I was
trying to find where the "//domain" was removed. it may be in the
agent_config_auto_link.c#370 where the addr is parsed.



> The current AMQP addressing draft *never* includes connection
> information in the AMQP address - because of exactly this kind of
> confusion. Given the direction of the current draft and the fact that
> it is still not final, I agree that dispatch should not modify the
> configured addresses at all.
>

I am fine with the "//" as long as it is not considered as connection
information :)


> I think Ganesh is already working on that so I'll stop confusing the
> issue.
>
>
>
Let us know if this is the case, that would be great!!

Reply via email to