get rid of keybase.socket and kbfs.socket

Having systemd open sockets for us is cool, but it was causing
unfortunate corner cases when the user switched back and forth between
KEYBASE_SYSTEMD=1 and KEYBASE_SYSTEMD=0 within the same session. This
mostly impacts just me, I think, cause I'm always toggling this setting,
but it hits me often enough that I don't want to risk it in the wild.

go/client/fork_server_nix.go

@@ -31,10 +31,7 @@ func spawnServer(g *libkb.GlobalContext, cl libkb.CommandLine, forkType keybase1
 	// add a systemd override file (see https://askubuntu.com/q/659267/73244).
 	if g.Env.WantsSystemd() {
 		g.Log.Info("Starting keybase.service.")
-		// Explicitly restarting the socket ensures it gets recreated, in case
-		// a non-systemd invocation of the service deleted it in the meantime
-		// without telling systemd it was "stopped".
-		startCmd := exec.Command("systemctl", "--user", "restart", "keybase.socket", "keybase.service")
+		startCmd := exec.Command("systemctl", "--user", "restart", "keybase.service")
 		startCmd.Stdout = os.Stderr
 		startCmd.Stderr = os.Stderr
 		err = startCmd.Run()

go/service/main.go

@@ -784,6 +784,11 @@ func (d *Service) lockPIDFile() (err error) {
 func (d *Service) ConfigRPCServer() (net.Listener, error) {
 	// First, check to see if we've been launched with socket activation by
 	// systemd. If so, the socket is already open. Otherwise open it ourselves.
+	// NOTE: We no longer configure our keybse.service and kbfs.service units
+	// to be socket-activated by default. It was causing too much trouble when
+	// non-systemd instances deleted the socket files. It's possible this issue
+	// will get fixed in future versions of systemd; see
+	// https://github.com/systemd/systemd/issues/7274.
 	listener, err := systemd.GetListenerFromEnvironment()
 	if err != nil {
 		d.G().Log.Error("unexpected error in GetListenerFromEnvironment: %#v", err)

go/systemd/systemd_linux.go

@@ -16,6 +16,11 @@ func IsRunningSystemd() bool {
 	return sdUtil.IsRunningSystemd()
 }
 
+// NOTE: We no longer configure our keybse.service and kbfs.service units to be
+// socket-activated by default. It was causing too much trouble when
+// non-systemd instances deleted the socket files. It's possible this issue
+// will get fixed in future versions of systemd; see
+// https://github.com/systemd/systemd/issues/7274.
 func IsSocketActivated() bool {
 	return (os.Getenv("LISTEN_FDS") != "")
 }

packaging/linux/run_keybase

@@ -64,9 +64,6 @@ start_systemd() {
   # If the services weren't started with systemd, we need to kill them.
   if ! was_started_with_systemd ; then
     kill_all
-    # Our non-systemd services tend to delete systemd's socket files. Make sure
-    # they get recreated.
-    systemctl --user stop keybase.socket kbfs.socket
   fi
   echo Starting via systemd.
   # The GUI needs access to the env var KEYBASE_START_UI. This is set to

packaging/linux/systemd/kbfs.service

@@ -1,8 +1,5 @@
 [Unit]
 Description=Keybase Filesystem service
-# kbfs.socket automatically knows to start kbfs.service, but for the reverse we
-# need this explicit Requires.
-Requires=kbfs.socket
 Wants=keybase.service
 
 [Service]

packaging/linux/systemd/keybase.service

@@ -1,8 +1,5 @@
 [Unit]
 Description=Keybase core service
-# keybase.socket automatically knows to start keybase.service, but for the
-# reverse we need this explicit Requires.
-Requires=keybase.socket
 
 [Service]
 # "notify" means we promise to call SdNotify() at the end of startup.