[rabbitmq-discuss] Delay delivering messages

[Michael Burns]

On Fri, Sep 17, 2010 at 4:13 PM, Vlad Alexandru Ionescu
<vlad at rabbitmq.com>wrote:

>
> Michael,
>
>
> So I understand that it is the consumer's connection that was hanging in
> there.
>
> I tried to reproduce your problem yesterday (killing off both consumer and
> producer apps), with no success.
> What version of Java are you using?
>

We're using Java 1.6.0_15. Our consumers are running on OS X 10.6.2


>
> By not having been shutdown cleanly... could it be possible, in your case,
> that the client has been disconnected from the network abruptly, in some
> way?
>

Dont think so - the main cause of the app not shutting cleanly is a
deployment script we have for our dev environments which kills our JVM if it
doesnt shutdown within a specified time period (because we are running in a
web container with multiple web apps sometimes it can take a couple of
seconds to successfully shutdown). Also it may be worth mentioning that our
consumer is on a different network to the RabbitMQ broker - the broker is
running at amazon on an EC2 instance, while the consumer runs on our
internal network.


>
> The only situation I can think of when you can have stray connections (and
> would not be a bug) would be the case when the socket has been broken in
> some way without doing a TCP FIN. If the app is killed, the underlying OS
> takes care of closing all its FDs, including all sockets that were held by
> the app - so that shouldn't cause stray connections. But, for example, if
> the client machine is physically disconnected from the network, the broker
> has no way of knowing that and expects the client to continue consuming (but
> it doesn't). Until the TCP socket times out (could take 15 mins), the
> message would remain blocked in the queue, as it is reserved for the
> disconnected client (which didn't receive the message entirely and/or didn't
> get to ack).
>

Yes this make sense - if I see it happening again I'll try and wait to see
if the connection eventually times out.

Thanks
Michael




>
>
> Rgds,
> Vlad.
>
>
>
> On 09/17/2010 03:56 PM, Michael Burns wrote:
>
> Hi,
>
> Just to add that I experienced the delay problem again - this time I
> checked the number of connections that were open and saw a stray connection
> was still hanging about. Upon manually killing this connection my message
> then came through.
> Any thoughts?
>
> (Again just to add that I'm not 100% sure our consumer app shutdown
> cleanly, it may have been killed before the connection could be closed our
> side)
>
> Thanks,
> Michael
>
> On Thu, Sep 16, 2010 at 1:39 PM, Michael Burns <michaelburns1 at gmail.com>wrote:
>
>> Hi Vlad,
>> We're using the Java client 1.8.1
>>
>>
>> Thanks,
>> Michael
>>
>>
>> On Thu, Sep 16, 2010 at 1:37 PM, Vlad Alexandru Ionescu <
>> vlad at rabbitmq.com> wrote:
>>
>>>
>>>
>>>> I've noticed sometimes that there are extra connections hanging about on
>>>> the RabbitMQ broker (using rabbitmqctl  list_connections).
>>>>
>>>
>>>  Michael, can you tell us which client(s) are you using?
>>>
>>> Thanks,
>>> Vlad.
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20100917/3ae3605d/attachment.htm>