Error handling in Camel
Error handling in Camel can roughly be separated into two distinct types:
- non transactional
Where non transactional is the most common type that is enabled out-of-the-box and handled by Camel itself. The transaction type is handled by a backing system such as a J2EE application server.
Using try ... catch ... finally
Related to error handling is the Try Catch Finally feature in Camel.
When does an error happen
An error happens when
any uncaught exception is thrown during routing and processing of messages within Camel
So think of this as a big exception interceptor that catches all exceptions and handles what to do.
By default Camel uses the non transaction type and orchestrates the error handling during processing and routing.
As there isn't a single error handling configuration that suites all uses cases, you should consider altering the default configurations to better suit you needs.
Camel 1.x default error handler
- redeliver up to 6 times
- pause 1 second between each redelivery attempt
- if all redelivery attempts failed then move exchange into the dead letter queue
- the default dead letter queue is a logger that logs the exchange at ERROR level
Dead Letter Queue (*)
A dead letter queue is like a black hole, it will consume the Exchange and the Exchange routing is ended with no indication that it failed.
This works great in the JMS Messaging world where we don't want a bad message to cause endless retries and causing the system to exhaust. The message is said to be poison and thus we want to move it to a dead letter queue so the system can continue to operate and work with the next message.
This default does not go well with other transports using in a request/reply messaging style. If the Exchange failed then the original caller want to be alter it failed.
So the bottom line is that you must configure and setup the error handling strategies that suits your business needs.
Camel 2.0 onwards default error handler
- no redeliveries
- no dead letter queue
- if the exchange failed an exception is thrown and propagated back to the original caller wrapped in a
Camel supports 2 scopes that is determined by the DSL in use:
Scope 2 takes precedence over scope 1
Scope 2 takes precedence over scope 1
When using XML DSL then scope 1 applies for all routes. Where as when using Java DSL then route 1 only applies for the given RouteBuilder instance. So if you have multiple RouteBuilder's then each route builder has its own scope 1.
If you want to share scope among RouteBuilder's you can use class inheritance and create a base class, and then extend this class for your RouteBuilder's and invoke the super.configure() method.
Mind that there was a bug in Camel that affected the scopes when using multiple RouteBuilder classes. See more details at CAMEL-5456.
How does the Dead Letter Channel error handler work
When Camel is started it will inspect the routes and weave in the error handling into the routing. With up to 3 supported scopes, the error handling can be quite complex. And on top of that you have inherited error handling and you can even configure Exception Clauses to handle specific exception types differently. So yes it's advanced but very powerful when you get the grip of it.
To keep things simple we first look at the basic concept how Camel orchestrates the redelivery attempt. At any given node in the route graph Camel intercepts the current Exchange being routed and wraps it with the Error Handler. This ensures that the Error Handler can kick in, just as the AOP around concept. If the exchange can be routed without any problems then it's forwarded to the next node in the route graph, But if there was an exception thrown, then the Error Handler kicks in and decides what to do.
An example illustrating this:
In this route we have 3 nodes (the dots) where the Error Handler is watching us (The AOP around stuff). So when an order arrives on the seda queue we consume it and send it to the validateOrder bean. In case the validation bean processed ok, we move on to the next node. In case the storeOrder bean failed and throws an exception it's caught by the Dead Letter Channel that decides what to do next. Either it does a:
- or move it to dead letter queue
It will continue to do redeliveries based on the policy configured. By default Dead Letter Channel will attempt at most 6 redeliveries with 1 second delay. So if the storeOrder bean did succeed at the 3rd attempt the routing will continue to the next node the confirmOrder bean. In case all redeliveries failed the Exchange is regarded as failed and is moved to the dead letter queue and the processing of this exchange stops. By default the dead letter queue is just a ERROR logger.
Camel leverages Spring transactions. Usually you can only use this with a limited number of transport types such as JMS or JDBC based, that yet again requires a transaction manager such as a Spring transaction, a J2EE server or a Message Broker.
How does it work
Camel does the same weaving as for the non-transactional type. The difference is that for transactional exchanges the Error Handler does not kick in. You can say the AOP around does not apply. Camel relies solely on the backing system to orchestrate the error handling. And as such the when the backing system does redeliver it will start all over again. For instance if the exchange was started by a JMS consumer then it's started again as the JMS message is rolled back on the JMS queue and Camel will re consume the JMS message again.
In Camel 2.0 we have empowered the TransactionErrorHandler to build on top of the same base that DefaultErrorHandler does. This allows you to use Camel redelivery with transactional routes as well. The Spring transaction manager is still in charge and have the last say. But you can use Camel to do some local redelivery, for instance to upload a file to a FTP server, in which Camel can do local redelivery. So this gives you the power from both worlds. In case Camel cannot redeliver the exchange will be failed and rolled back. By default the TransactionErrorHandler does not attempt any local redeliveries. You have to configure it to do so, for instance to set a maximum redelivers to a number > 0.
See Transactional Client for more.