[rabbitmq-discuss] Strange reaction to publishing message with invalid expiration.

[James Gardner]

Simon,
That helps me understand - thank you!
Cheers, James


On 03/19/2013 10:13 AM, Simon MacMullen wrote:
> Hi James. Yes, the basicPublish() invocation is not guaranteed to 
> throw an exception, since it's not synchronous - the message may not 
> have reached the server (or even left the client) by the time the 
> method returns. An exception will get thrown at some point later. If 
> you need to ensure that you find out about errors immediately you can 
> close the channel (and create a new one), invoke any synchronous AMQP 
> method (almost anything apart from basic.publish), or wait for confirms.
>
> Cheers, Simon
>
> On 19/03/13 14:59, James Gardner wrote:
>> Didn't get a response to email below, sending again in hopes someone can
>> answer...
>>
>> ==============
>>
>> Ok thanks, good to know. If that's how AMQP handles errors, that part's
>> fine, and I'm used to PHP extensions not dealing with things correctly
>> :), but I don't appear to be getting an exception when my Java code
>> reaches:
>> channel.basicPublish("e.test", "", props, "Hi!!!".getBytes());
>> That statement, along with everything else (including the call to
>> consume, which *does* throw an exception just fine) is inside a try
>> {...} catch (Exception e) {...} .
>> If you say I should be getting an exception, then I cannot explain why I
>> am not.
>> Here, I'll just include the code. All of this is just inside a main
>> function in a Test class (I took out the consume part):
>>
>> try {
>>              ConnectionFactory factory = new ConnectionFactory();
>> factory.setUri("amqp://user:pwd@amqp1.ourserver.gov/ourvhost");
>>              Connection conn = factory.newConnection();
>>
>>              Channel channel = conn.createChannel();
>>
>>              BasicProperties props = new BasicProperties
>>                                      .Builder()
>>                                      .expiration("trete")
>>                                      .build();
>>
>>              channel.basicPublish("e.test", "", props, 
>> "Hi!!!".getBytes());
>>
>>
>>          }
>>          catch (Exception e) {
>>              System.out.println("Threw exception: ");
>>              e.printStackTrace(System.out);
>>          }
>>
>> My Java is a little rusty so forgive me if I am missing something.
>> Am I supposed to get an IOException if the publish fails due to the bad
>> header, or only if the channel had been closed before I do the publish
>> or if I tried to use the channel again after the bad publish? The
>> Javadocs seem vague on this.
>>
>> - James
>>
>>
>> On 03/12/2013 11:12 AM, Simon MacMullen wrote:
>>> Well, the channel is getting closed with an error message (same as it
>>> would be for an illegal user_id in fact). That's AMQP's approach to
>>> error handling.
>>>
>>> You should be able to see an IOException getting thrown in the Java
>>> client, with a more descriptive cause property, or you can register a
>>> ShutdownListener. I have no idea how the PHP client signals errors I'm
>>> afraid.
>>>
>>> Cheers, Simon
>>>
>>> On 12/03/13 14:49, James Gardner wrote:
>>>>
>>>> Hi,
>>>> I found a behavior which, to me at least, acts more like a bug than a
>>>> feature...
>>>> It seems that when you publish a message with an invalid expiration
>>>> string (eg. "blah"), it renders that channel inoperable for consume
>>>> operations afterwards. I would think it should just discard the 
>>>> message.
>>>> In the official Java client, something closes the channel, although I
>>>> don't get any exception until I try to use it for the ensuing consume
>>>> operation.
>>>> In the PHP AMQP client (this one:
>>>> http://www.php.net/manual/en/book.amqp.php), the channel stays open 
>>>> but
>>>> when you execute $queue->consume(...) and look in the management
>>>> interface you can see it has not registered as a consumer on the queue
>>>> of interest.
>>>> Given the commonalities, I assume these are different client
>>>> interpretations of the same broker behavior. I am not using 
>>>> transactions
>>>> or publisher confirms or anything fancy. I am less familiar with the
>>>> Java client so perhaps I am missing some sort of error handling?
>>>>
>>>> RabbitMQ version: RabbitMQ 3.0.3, Erlang R14B04 on RHEL 6.4
>>>> PHP AMQP version is 1.0.1 (admittedly a little outdated)
>>>> Java client is version: 3.0.3
>>>>
>>>> I can provide code for either client if you would like but I 
>>>> imagine you
>>>> can replicate it easily just by establishing a connection, then a 
>>>> single
>>>> channel, then publishing the 'bad' msg to a topic exchange, then 
>>>> trying
>>>> to consume off a queue bound to that exchange with '#'.
>>>> When you change expiration to a valid string integer, everything works
>>>> as you would wish.
>>>>
>>>> Of course one answer, as the doctor joke goes, is to not 'move your 
>>>> arm
>>>> like that'. Which is fine so long as there's a choice, but if I were
>>>> creating messages and taking the expiration from a source outside my
>>>> program (eg. database), this error might occur at some point if the 
>>>> data
>>>> were bad. I could/should(?) sanitize/verify first of course, the point
>>>> is more that I thought the typical broker reaction to illegal data was
>>>> to deal with it 'silently' (as it does with an illegal user_id, for
>>>> example) rather than effectively make the channel unusable.
>>>> Is there a different way I (or the client code) should be handling 
>>>> this
>>>> or is this an unexpected broker behavior?
>>>> Thanks,
>>>>
>>>> James Gardner
>>>> NWS - Internet Dissemination System
>>>>
>>>>
>>>> _______________________________________________
>>>> rabbitmq-discuss mailing list
>>>> rabbitmq-discuss at lists.rabbitmq.com
>>>> https://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss
>>>
>>>
>>
>
>