Chapter 6. Running a Camel Service on Spring Boot with XA Transactions

The Spring Boot Narayana recovery controller allows to gracefully handle the scaling down phase of a StatefulSet by cleaning pending transactions before termination. If a scaling down operation is executed and the pod is not clean after termination, the previous number of replicas is restored, hence effectively canceling the scaling down operation.

All pods of the StatefulSet require access to a shared volume that is used to store the termination status of each pod belonging to the StatefulSet. The pod-0 of the StatefulSet periodically checks the status and scale the StatefulSet to the right size if there’s a mismatch.

In order for the recovery controller to work, edit permissions on the current namespace are required (role binding is included in the set of resources published to OpenShift). The recovery controller can be disabled using the CLUSTER_RECOVERY_ENABLED environment variable. In this case, no special permissions are required on the service account but any scale down operation may leave pending transactions on the terminated pod without notice.

Following example shows how to configure Narayana to work on OpenShift with the recovery controller. This is a sample application.properties file:

# You need to replace the following options in the Kubernetes yaml descriptor, see below
cluster.nodename=1
cluster.base-dir=./target/tx

# TX manager
spring.jta.transaction-manager-id=${cluster.nodename}
spring.jta.log-dir=${cluster.base-dir}/store/${cluster.nodename}

# Narayana recovery settings
snowdrop.narayana.openshift.recovery.enabled=true
snowdrop.narayana.openshift.recovery.current-pod-name=${cluster.nodename}
# You must enable resource filtering in order to inject the Maven artifactId
snowdrop.narayana.openshift.recovery.statefulset=${project.artifactId}
snowdrop.narayana.openshift.recovery.status-dir=${cluster.base-dir}/status

You need a shared volume to store both transactions and information related to termination. It can be mounted in the StatefulSet yaml descriptor:

apiVersion: apps/v1beta1
kind: StatefulSet
#...
spec:
#...
  template:
#...
    spec:
      containers:
      - env:
        - name: CLUSTER_BASE_DIR
          value: /var/transaction/data
          # Override CLUSTER_NODENAME with Kubernetes Downward API (to use `pod-0`, `pod-1` etc. as tx manager id)
        - name: CLUSTER_NODENAME
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: metadata.name
#...
        volumeMounts:
        - mountPath: /var/transaction/data
          name: the-name-of-the-shared-volume
#...

If Camel is found in the Spring Boot application context, the Camel context is automatically stopped before flushing all pending transactions.

Following workflow shows how to test the failed XA transactions: