There are several types of error conditions. In this document, a class of error condition called failures is introduced, distinct from faults, and describe how failures are caught and handled by the process engine.
A service returns a fault in response to a request it cannot process. A process may also raise a fault internally when it encounters a terminal error condition, e.g. a faulty expression or false join condition. In addition, processes may raise faults in order to terminate normal processing.
In contrast, failures are non-terminal error conditions that do not affect the normal flow of the process. The process definition is kept simple and straightforward by delegating failure handling to the process engine and administrator.
For example, when the process is unable to perform DNS resolution to determine the service endpoint, it generates a failure. An administrator can fix the DNS server and tell the process engine to retry the activity. Had the DNS error been reported as a fault, the process would either terminate or require complex fault handling and recovery logic to proceed past this point of failure.
In short, failures shields the process from common, non-terminal error conditions while retaining simple and straightforward process definitions that do not need to account for these error conditions.
Currently, the Invoke activity is the only activity that supports failure handling and recovery. The mechanism is identical for all other activities that may support failure handling and recovery in the future.
In case of the Invoke activity, a failure condition is triggered by the integration layer, in lieu of a response or fault message. The Invoke activity consults its failure handling policy (specified here) and decides how to respond.
Set faultOnFailure to yes, if you want the activity to throw a fault on failure. All other failure handling settings are ignored. The activity will throw the activityFailure fault.
The activityFailure fault is a standard fault, so you can use the exitOnStandardFault attribute to control whether the process exits immediately, or throws a fault in the enclosing scope.
Set retryFor to a positive integer if you want the activity to attempt self-recovery and retry up to that number of times. Set retryDelay to a reasonable time delay (specified in seconds) between retries. For example, if you set retryFor=2, retryDelay=30, the activity will retry after 30 and 60 seconds, for a total of three attempts, before entering activity recovery mode.
If the activity retries and succeeds, it will complete successfully as if no failure occurred. Of course, the activity may retry and fault, e.g. if the invoked service returns a fault. If the activity has exhausted all retry attempts, it enters activity recovery mode. By default retryFor is zero, and the activity enters recovery mode after the first failure.
When in recovery mode, you can recover the activity in one of three ways:
Activity recovery is performed individually for each activity instance, and does not affect other activities executing in the same process. While the activity is in the FAILURE state, the process instance remains in the ACTIVE state and may execute other activities from parallel flows and event handlers.
Use the failureHandling extensibility element defined in the namespace http://ode.apache.org/activityRecovery. The structure of the failureHandling element is:
<ext:failureHandlingxmlns:ext="http://ode.apache.org/activityRecovery"><ext:faultOnFailure> _boolean_ </ext:faultOnFailure><ext:retryFor> _integer_ </ext:retryFor><ext:retryDelay> _integer_ </ext:retryDelay></ext:failureHandling>
The faultOnFailure, retryFor and retryDelay elements are optional. The default values are false for faultOnFailure, and zero for retryFor and retryDelay. An activity that does not specify failure handling using this extensibility element, inherits the failure handling policy of its parent activity, recursively up to the top-level activity of the process. You can use inheritance to specify the failure handling policy of a set of activities, or all activities in the process, using a single failureHandling extensibility element.
Note that due to this behavior, if activity S specifies failure handling with the values retryFor=2, retryDelay=60, and has a child activity R that specifies failure handling with the values retryFor=3, the retryDelay value for the child activity R is 0, and not 60. Using the failureHandling element without specifying one of its value elements will apply the default value for that element.
A simple invoke with the ext:failureHandling extension:
And a sequence activity that converts failures into faults:
<bpel:sequence><ext:failureHandlingxmlns:ext="http://ode.apache.org/activityRecovery"><ext:faultOnFailure>true</ext:faultOnFailure></ext:failureHandling> ... <bpel:invokeinputVariable="myRequest"operation="foo"outputVariable="aResponse"partnerLink="myPartner"portType="spt:SomePortType"><bpel:catchAll> ... </bpel:catchAll></bpel:invoke></bpel:sequence>
The process instance management provides the following information:
Use the recoverActivity operation to perform a recovery action on an activity in recovery mode. The operation requires the process instance ID, the activity instance ID and the recovery action to perform (one of retry, fault or cancel).
It can be also determined when failure or recovery occurred for a given activity instance from the execution log.