Robert Szczesiak created CAMEL-14789:
----------------------------------------
Summary: Automatic recovery of temporary reply queue is not
handled correctly
Key: CAMEL-14789
URL: https://issues.apache.org/jira/browse/CAMEL-14789
Project: Camel
Issue Type: Bug
Components: camel-rabbitmq
Affects Versions: 2.25.0
Environment: * RabbitMQ server 3.8.3
* Erlang 22.3
Reporter: Robert Szczesiak
Fix For: 2.25.1
When Remote Procedure Call (RPC) communication pattern is used RabbitMQ Camel
component creates server-named auto-deleted temporary reply queue. Next, the
queue is bound to the exchange with routing key equal to queue name. Likewise,
ReplyManager's replyTo field is set with the value of the queue name. Reply
queue created in this manner is then reused for subsequent RPC requests and
message property rabbitmq.REPLY_TO is copied from the value stored previously
by ReplyManager.
When some network error suddenly appears causing connection failure, RabbitMQ's
automatic recovery kicks in and tries to recover affected entities (assuming
the connection was created with automatic recovery enabled, which is default.).
As temporary quueues are auto-deleted, during recovery process a new temporary
queue is created which has a new name that differs from the original one and
here is where the problem begins.
Creation of the new temporary queue is NOT detected by ReplyManager and
therefore replyTo property is NOT updated. Also, routing key no longer matches
queue name. This causes a problem when some implementations of RabbitMQ client,
like. Spring AMQP, are used server-site. RPC service receives our request,
processes it, and replies to default exchange with routing key equal to
rabbitmq.REPLY_TO sent in our request. RPC service provider perceives no
problem as response is sent successfully but RabbitMQ Camel Component keeps
awaiting for the response to arrive to the original temporary queue which no
longer exists due to connection failure and recovery. Eventually, Camel throws
ExchangeTimedOutException.
Example:
# After automatic recovery, a RPC request is sent to example-exchange with
rabitmq.REPLY_TO=amq.gen-0lLvpnj4ZMlkhxZIcCPVpA
# After 20 seconds org.apache.camel.ExchangeTimedOutException is thrown:
{code:java}
The OUT message was not received within: 20000 millis due reply message with
correlationID:
Camel-ID-hostname-y579vjqenbwmcd1bpqmd5phtd-1583401848872-0-1690397 not
received on destination: amq.gen-0lLvpnj4ZMlkhxZIcCPVpA.
Exchange[ID-hostname-y579vjqenbwmcd1bpqmd5phtd-1583401848872-0-1690391]{code}
# Using RabbitMQ Management WebApp (rabbitmqctl) we check example-exchange's
bindings and see that routing key amq.gen-0lLvpnj4ZMlkhxZIcCPVpA now
corresponds to queue amq.gen-zaRCP-p-JbXeSzJmzSp83g
*Proposed solution* is to add QueueRecoveryListener to notify when temporary
queue name changes due to recovery. On event, replyTo field will be updated
with the new temporary queue name and the queue will be rebound to the exchange
so that routing key matches queue name again. The change will be made to
org.apache.camel.component.rabbitmq.reply#createListenerContainer.
The fix is ready to be delivered in pull-request/patch.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
