concurrent skaffold dev processes fail with RPC port conflicts

[matti]

start two skaffold dev processes concurrently:

skaffold dev -f first.yaml & 
skaffold dev -f second.yaml &

process1: WARN[0000] port 50051 for gRPC server already in use: using 50052 instead
process2: WARN[0000] port 50051 for gRPC server already in use: using 50052 instead
initializing api server: starting gRPC server: creating listener: listen tcp 127.0.0.1:50052: bind: address already in use

the other one will exit because it fails to bind.

workaround:

for yaml in first.yaml second.yaml; do
  (
    while true; do
      skaffold dev -f $yaml || true
      echo "don't give up, skaffold!"
    done
  ) &
done

[briandealwis]

@matti you can specify a port with the --rpc-port and --rpc-http-port command-line options.

We try to select a port near the default port, but there is a race condition between checking for the availability of a port vs actually binding to that port. We should probably have some retry logic that will bind to a random port via port 0 if the original attempt fails.

[mecampbellsoup]

Still experiencing this issue with v1.22.0:

(⎈ minikube:tenant-ad02d6c1b)➜  dev git:(develop) ✗ skaffold dev -f skaffold-k8s-billing.yaml
initializing api server: starting gRPC server: creating listener: listen tcp 127.0.0.1:50060: bind: address already in use
(⎈ minikube:tenant-ad02d6c1b)➜  dev git:(develop) ✗ skaffold version
v1.22.0