We ran into a situation where the multiheaded registry and auth servers were trying to connect to managers that were no longer running. After several days the manager process was consuming ~13 GB of memory and responding very slowly (if at all) to new connections. It would appear that there is a slow memory leak when the multiheaded servers fail to connect. Could be netty buffers that are not getting released, for example.
We ran into a situation where the multiheaded registry and auth servers were trying to connect to managers that were no longer running. After several days the manager process was consuming ~13 GB of memory and responding very slowly (if at all) to new connections. It would appear that there is a slow memory leak when the multiheaded servers fail to connect. Could be netty buffers that are not getting released, for example.