Once we enable linkerd on production, linkerd destination pods were restarting due to OOM because of memory limit… This increased the open connection and sockets on the application where proxies were enabled. This in-turn increased the actual application pod memory usage and the application pods started to restart thus affecting production.
Does restarting linkerd destination control plane pods has any effect on data plane proxies ?
It should not, except to the extent that during the period where the destination control plane is unavailable, existing proxies won’t get updates from service discovery (so may try to establish connections to non-existent endpoints, or may not know about new endpoints); and new proxies won’t be able to function. Once the destination service has recovered, the proxies should recover as well.