Open Issues
Learn about the open issues in Address Pool Manager 3.5.0
-
In a multiple geography deployment, if a node is lost in a workload cluster and the microservices that were running on that node cannot be scheduled on the remaining nodes in the workload cluster (for example there are insufficient resources to satisfy the resource requests of the evicted microservices), the microservices will be scheduled on the peer workload cluster. This creates a partial switchover scenario. APM requires that all microservices that exist on only one workload cluster at a time switch over together. APM in a partial switchover state will not function properly.
Workaround:
In the event of a partial switchover, initiate a graceful switchover of APM to the peer workload cluster. This forces all microservices to run on the same workload cluster.
-
On rare occasions, the
show apm entity id <id> pool-domain <poolDomainName>command returns the response:Error retrieving prefix utilization data for entity <entityId>. PR1895333 -
Deleting an
entity-matchentry does not completely clean up theshow apm entityoutput. PR1874241 -
An in-service upgrade of the bbe-observer triggers a rollout. As part of an in-service upgrade of the observer microservice, it may appear as if all APM microservices are being upgraded. The list of container images loaded or pushed in the
rolloutoutput indicate that all the microservices are being upgraded, but only the observer microservice is upgraded. Other container images that are being loaded or pushed are not upgraded. PR1879715 -
Reverting network load balancer (MetalLB) annotations followed by performing a rollout does not reset the external IP address for the APMi.
Workaround:
When the external address of the APMi that is bound to a particular IPAddressPool through the network load balancer annotations needs to be reverted to use an autoAssign IPAddressPool by removing the annotations, a
stopcommand and then arolloutcommand of APM must be performed.