The way the Kubernetes plugin tracks pod usage, a slot is registered as soon as a Planned Node is created:

• https://github.com/jenkinsci/kubernetes-plugin/blob/3883.v4d70a_a_a_df034/src/main/java/org/csanchez/jenkins/plugins/kubernetes/KubernetesCloud.java#L547

And unregistered only when a Node is deleted:

• https://github.com/jenkinsci/kubernetes-plugin/blob/3883.v4d70a_a_a_df034/src/main/java/org/csanchez/jenkins/plugins/kubernetes/KubernetesProvisioningLimits.java#L139-L149

There are however cases where a Planned Node is indeed created but the resulting Node never created. For example:

• https://github.com/jenkinsci/kubernetes-plugin/blob/3883.v4d70a_a_a_df034/src/main/java/org/csanchez/jenkins/plugins/kubernetes/KubernetesCloud.java#L547

• https://github.com/jenkinsci/jenkins/blob/jenkins-2.387.1/core/src/main/java/hudson/slaves/NodeProvisioner.java#L231-L240
• https://github.com/jenkinsci/kubernetes-plugin/blob/9c1d58d9badd988622b4772632037126176e1ef0/src/main/java/org/csanchez/jenkins/plugins/kubernetes/KubernetesCloud.java#L613-L630

Which could lead to situation where slots are leaked and the provisioning limit is reached while in reality it is not.