Feature #19166
Updated by Tom Clegg about 2 years ago
Unlike Since the arvados-dispatch-cloud case, the dispatcher doesn't know which HPC compute what node something will be run the container, and the HPC compute node isn't necessarily even reachable from controller. To work around this, on, we will make an initial connection in the opposite direction and set up a tunnel. * probably need crunch-run connects to new controller API arvados/v1/containers/{uuid}/gateway_tunnel, authenticated using the container key (GatewayAuthSecret) * controller registers figure out its own internalURL as the container’s GatewayAddress, hostname or IP address and uses update the tunnel to route incoming container_ssh connections to crunch-run through the tunnel container record accordingly. * there can be multiple controller hosts/processes; the container_ssh API on controller A will sometimes need to proxy through the same API on controller B