Since few weeks, we remove component server from Cloud engine deployment and we are using only two instances for the all cluster (deployed in the multi-tenant namespace).

With the default ressources configuration, 1G of memory, the number of restart of this service due to OOMKilled is big - like almost 100 restart for a couple of days.

I bump to 2g for the request memory but we still have the same pattern, even if we have less restart.

You can see in the screenshot the behavior of the service in terms of memory consumption.

I guess a Thread dump could be a first option to analyze the memory usage before the restart of the service.