gyfora commented on code in PR #165:
URL:
https://github.com/apache/flink-kubernetes-operator/pull/165#discussion_r850305005
##########
flink-kubernetes-operator/src/main/java/org/apache/flink/kubernetes/operator/reconciler/deployment/ApplicationReconciler.java:
##########
@@ -102,36 +111,84 @@ public void reconcile(FlinkDeployment flinkApp, Context
context, Configuration e
}
if (currentJobState == JobState.SUSPENDED && desiredJobState ==
JobState.RUNNING) {
if (upgradeMode == UpgradeMode.STATELESS) {
- deployFlinkJob(flinkApp, effectiveConfig,
Optional.empty());
- } else if (upgradeMode == UpgradeMode.LAST_STATE
- || upgradeMode == UpgradeMode.SAVEPOINT) {
- restoreFromLastSavepoint(flinkApp, effectiveConfig);
+ deployFlinkJob(currentJobSpec, status, effectiveConfig,
Optional.empty());
+ } else {
+ restoreFromLastSavepoint(currentJobSpec, status,
effectiveConfig);
}
stateAfterReconcile = JobState.RUNNING;
}
- IngressUtils.updateIngressRules(flinkApp, effectiveConfig,
kubernetesClient);
+ IngressUtils.updateIngressRules(
+ deployMeta, currentDeploySpec, effectiveConfig,
kubernetesClient);
ReconciliationUtils.updateForSpecReconciliationSuccess(flinkApp,
stateAfterReconcile);
- } else if (SavepointUtils.shouldTriggerSavepoint(flinkApp) &&
isJobRunning(flinkApp)) {
+ } else if (ReconciliationUtils.shouldRollBack(reconciliationStatus,
effectiveConfig)) {
+ rollbackApplication(flinkApp);
+ } else if (SavepointUtils.shouldTriggerSavepoint(currentJobSpec,
status)
+ && isJobRunning(status)) {
triggerSavepoint(flinkApp, effectiveConfig);
ReconciliationUtils.updateSavepointReconciliationSuccess(flinkApp);
+ } else {
+ LOG.info("Deployment is fully reconciled, nothing to do.");
}
}
+ private void rollbackApplication(FlinkDeployment flinkApp) throws
Exception {
+ ReconciliationStatus reconciliationStatus =
flinkApp.getStatus().getReconciliationStatus();
+
+ if (reconciliationStatus.getState() !=
ReconciliationStatus.State.ROLLING_BACK) {
+ LOG.warn("Preparing to roll back to last stable spec.");
+ if (flinkApp.getStatus().getError() == null) {
+ flinkApp.getStatus()
+ .setError(
+ "Deployment is not ready within the configured
timeout, rolling-back.");
+ }
+
reconciliationStatus.setState(ReconciliationStatus.State.ROLLING_BACK);
+ return;
+ }
+
+ LOG.warn("Executing roll-back operation");
+
+ FlinkDeploymentSpec rollbackSpec =
reconciliationStatus.deserializeLastStableSpec();
+ Configuration rollbackConfig =
+ FlinkUtils.getEffectiveConfig(flinkApp.getMetadata(),
rollbackSpec, defaultConfig);
+
+ UpgradeMode upgradeMode = flinkApp.getSpec().getJob().getUpgradeMode();
+
+ suspendJob(
Review Comment:
I am afraid it is not that simple and it is by design like this. Let me try
to explain.
I think we earlier decided against making changes to the spec itself as it
breaks the idempotency of the operator logic. If we simply change the spec then
an external process might simply reapply what was previous rolled back causing
a loop.
This is also the reason why we do not upgrade the `lastReconciledSpec`. That
field shows the last spec that the user have sent in that was reconciled by the
controller. Even if it was rolled back later that does not change.
The ReconciliationStatus.State shows what actually happened to the
lastReconciledSpec: DEPLOYED means it is running on the cluster,
ROLLING_BACK/ROLLED_BACK actually signal that it was reverted after it was
DEPLOYED.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
