OWLS-87753 - Fix to start all managed server pods when watch event notifications not received in jumbo k8s cluster #2188
Conversation
…t notifications are not received
| return createReadAsyncStep(getName(), getNamespace(), getDomainUid(), resumeIfReady(callback)); | ||
| } else { | ||
| DomainPresenceInfo info = packet.getSpi(DomainPresenceInfo.class); | ||
| return createReadAsyncStep(resourceName, Optional.ofNullable(info).map(i -> i.getNamespace()).orElse(null), |
There was a problem hiding this comment.
do we need to add check to ensure resourceName is not null?
There was a problem hiding this comment.
The resourceName and initialResource are final variables and are initialized in the constructor. I'm assuming that either initialResource or resourceName will be non-null for the resource that needs to be checked. I can add a non-null check to be safe but I wasn't sure if that's going to be necessary. Thanks.
|
Will there be new tests added for this change? If not in this PR, please file a jira so we won't forget to add tests. Thanks |
|
| packet.clone(), | ||
| null); | ||
| } | ||
|
|
||
| private Step createReadAndIfReadyCheckStep(Callback callback) { | ||
| return createReadAsyncStep(getName(), getNamespace(), getDomainUid(), resumeIfReady(callback)); | ||
| private Step createReadAndIfReadyCheckStep(Callback callback, Packet packet) { |
There was a problem hiding this comment.
This doesn't follow our Step/Packet pattern. The Step is typically stateless or, at least, can be used for many different calls. The Packet is given on the Step.apply call. The logic in the else should happen in the Step.apply().
There was a problem hiding this comment.
I have made the suggested change, please let me know if I missed something. Thanks.
* Update for 3.1.3 * keep namespaces processed in batches (#2189) * OWLS-87753 - Fix to start all managed server pods when watch event notifications not received in jumbo k8s cluster (#2188) * Fix for OWLS_87753 to start managed server waiters in case watch event notifications are not received * Check for null initialResource * PR review comment - added null check for resourceName * Address PR review comment. * Release note update * Spelling and grammar updates Co-authored-by: Dongbo Xiao <dongbo.xiao@oracle.com> Co-authored-by: Anil Kedia <37935279+ankedia@users.noreply.github.com>
…tifications not received in jumbo k8s cluster (oracle#2188) * Fix for OWLS_87753 to start managed server waiters in case watch event notifications are not received * Check for null initialResource * PR review comment - added null check for resourceName * Address PR review comment.
…tifications not received in jumbo k8s cluster (#2188) * Fix for OWLS_87753 to start managed server waiters in case watch event notifications are not received * Check for null initialResource * PR review comment - added null check for resourceName * Address PR review comment.
OWLS-87753- Fix for starting managed servers when watch event notifications are missed.
This change creates waiters for each managed server pod in ManagedServerUpIterator step. These fibers are started in parallel with fibers that schedule and actually start the managed server pods. The waiter fibers invoke WaitForReady step for the corresponding managed server. At the time when WaitFotReady is invoked, managed server pods would not have been created. Hence it passes resourceName to the WaitForReady step instead of the initial resource. The managed server pod state is updated periodically in the DomainPresenceInfo when the callback onSuccess is invoked (every 5 seconds) in waitForReady step. The createReadAndIfReadyCheckStep uses domain-id and namespace from the packet instead of getting it from the initial resource.
The link for integration test results is at - https://build.weblogick8s.org:8443/job/weblogic-kubernetes-operator-kind-new/4024/ . There are 5 failures but they seem unrelated to this change.