Cloud endpoint management leaking memory?

Question

I have a strange problem with GAE Java. There are two instances with basic scaling for the version I am using, with one being used and the other idling, from what I can see in the log. Response times are fine. I can see that my idle instance did not receive any requests for the last hour. Strangely, on the idle instance the memory usage goes up constantly at around a 2MB/minute. For the last hour.

In production (total of both instances) this leads to a sawtooth type memory usage graph, where instances get killed regularely.

I now commented out endpoint service configs:

<filter>
    <filter-name>endpoints-api-configuration</filter-name>
    <filter-class>com.google.api.control.ServiceManagementConfigFilter</filter-class>
</filter>

&lt;!&ndash; Add a filter that performs Endpoints logging and monitoring. &ndash;&gt;
<filter>
    <filter-name>endpoints-api-controller</filter-name>
    <filter-class>com.google.api.control.extensions.appengine.GoogleAppEngineControlFilter</filter-class>
    <init-param>
        <param-name>endpoints.projectId</param-name>
        <param-value>${appengine.project.id}</param-value>
    </init-param>
    <init-param>
        <param-name>endpoints.serviceName</param-name>
        <param-value>${appengine.project.id}.appspot.com</param-value>
    </init-param>
</filter>

<filter-mapping>
    <filter-name>endpoints-api-configuration</filter-name>
    <servlet-name>EndpointsServlet</servlet-name>
</filter-mapping>

<filter-mapping>
    <filter-name>endpoints-api-controller</filter-name>
    <servlet-name>EndpointsServlet</servlet-name>
</filter-mapping>

And it seems to not leak memory anymore (this is on B4):

What could be the problem here? I guess a B4 instance should be enough to use service management?