app_container_tag is greater than the number of component instances

the way to reproduce this issue

• shutdown a nodemanager where a container is running.
• kill the container process.
• after a while, the container is released by HeartbeatMonitor.
• Slider AM request a new container to replace that.
• a new container container is allocated to another nodemanager.

but, app_container_tag assigned to a new container is greater than the number of component instances (configured by yarn.component.instances)
i think it does not clear app_container_tag when container is released by HeartbeatMonitor