Go App Served via SSH
bobhenkel opened this issue · 3 comments
I'm starting to experiment where the edges of an unikernel environment are. I didn't expect this to work and it doesn't. I wrote a simple ssh server in go that in theory serves out a simple app, it doesn't allow the user to execute any commands, only to hit the ssh server and be served content much like a http hello world REST API. The app listens for ssh connections on port 23234. When it's running you can connect to it with ssh localhost -p23234 and it will create a PTY and print some text back.
I'm curious if it's possible with a nanos instance? If not great no big deal, I'll move onto using nanos in a more typical manner. If it can work what's needed? Thanks!
When NOT ran in a nanos unikernel this is the output when connecting to the the ssh server:
Hello, world!
Term: xterm-256color
PTY: /dev/pts/5
Connection to localhost closed.
When ran in a nanos unikernel I get this error when I deploy it to a google cloud instance:
PTY allocation request failed on channel 0
Requires an active PTY
Connection to 104.999.999.63 closed.
Here's the code:
package main
import (
"context"
"errors"
"net"
"os"
"os/signal"
"syscall"
"time"
"github.com/charmbracelet/log"
"github.com/charmbracelet/ssh"
"github.com/charmbracelet/wish"
"github.com/charmbracelet/wish/activeterm"
"github.com/charmbracelet/wish/logging"
)
const (
host = ""
port = "23234"
)
func main() {
srv, err := wish.NewServer(
wish.WithAddress(net.JoinHostPort(host, port)),
// wish.WithHostKeyPath(".ssh/id_ed25519"),
// Wish can allocate a PTY per user session.
ssh.AllocatePty(),
wish.WithMiddleware(
func(next ssh.Handler) ssh.Handler {
return func(sess ssh.Session) {
pty, _, _ := sess.Pty()
wish.Printf(sess, "Hello, world!\r\n")
wish.Printf(sess, "Term: "+pty.Term+"\r\n")
wish.Printf(sess, "PTY: "+pty.Slave.Name()+"\r\n")
next(sess)
}
},
activeterm.Middleware(),
logging.Middleware(),
),
)
if err != nil {
log.Error("Could not start server", "error", err)
}
done := make(chan os.Signal, 1)
signal.Notify(done, os.Interrupt, syscall.SIGINT, syscall.SIGTERM)
go func() {
log.Info("Starting SSH server", "host", host, "port", port)
if err = srv.ListenAndServe(); err != nil && !errors.Is(err, ssh.ErrServerClosed) {
log.Error("Could not start server", "error", err)
done <- nil
}
}()
<-done
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer func() { cancel() }()
log.Info("Stopping SSH server")
if err := srv.Shutdown(ctx); err != nil && !errors.Is(err, ssh.ErrServerClosed) {
log.Error("Could not stop server", "error", err)
}
}
it doesn't allow the user to execute any commands, only to hit the ssh server and be served content much like a http hello world REST API
you don't need a "real" PTY then.
Can you try and check if the following changes work for you?
- comment/delete
ssh.AllocatePty(),or change it tossh.EmulatePty(), - remove
wish.Printf(sess, "PTY: "+pty.Slave.Name()+"\r\n")since there are no fd allocated or change it to smth likeif !pty.IsZero() { wish.Printf(sess, "PTY: "+pty.Slave.Name()+"\r\n") }
Thanks for your time, it's appreciated! Seems so obvious now that someone showed me the solution 😄. For anyone else that is trying to use the Charm modules to make ssh apps like the Wish module here's a more interesting app from the Wish module examples that just works out of the box. I see myself leaning into Nano unikernels more for some of my services and apps.
package main
// An example Bubble Tea server. This will put an ssh session into alt screen
// and continually print up to date terminal information.
import (
"context"
"errors"
"fmt"
"net"
"os"
"os/signal"
"syscall"
"time"
tea "github.com/charmbracelet/bubbletea"
"github.com/charmbracelet/lipgloss"
"github.com/charmbracelet/log"
"github.com/charmbracelet/ssh"
"github.com/charmbracelet/wish"
"github.com/charmbracelet/wish/activeterm"
"github.com/charmbracelet/wish/bubbletea"
"github.com/charmbracelet/wish/logging"
)
const (
host = ""
port = "23234"
)
func main() {
s, err := wish.NewServer(
wish.WithAddress(net.JoinHostPort(host, port)),
wish.WithMiddleware(
bubbletea.Middleware(teaHandler),
activeterm.Middleware(), // Bubble Tea apps usually require a PTY.
logging.Middleware(),
),
)
if err != nil {
log.Error("Could not start server", "error", err)
}
done := make(chan os.Signal, 1)
signal.Notify(done, os.Interrupt, syscall.SIGINT, syscall.SIGTERM)
log.Info("Starting SSH server", "host", host, "port", port)
go func() {
if err = s.ListenAndServe(); err != nil && !errors.Is(err, ssh.ErrServerClosed) {
log.Error("Could not start server", "error", err)
done <- nil
}
}()
<-done
log.Info("Stopping SSH server")
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer func() { cancel() }()
if err := s.Shutdown(ctx); err != nil && !errors.Is(err, ssh.ErrServerClosed) {
log.Error("Could not stop server", "error", err)
}
}
// You can wire any Bubble Tea model up to the middleware with a function that
// handles the incoming ssh.Session. Here we just grab the terminal info and
// pass it to the new model. You can also return tea.ProgramOptions (such as
// tea.WithAltScreen) on a session by session basis.
func teaHandler(s ssh.Session) (tea.Model, []tea.ProgramOption) {
// This should never fail, as we are using the activeterm middleware.
pty, _, _ := s.Pty()
// When running a Bubble Tea app over SSH, you shouldn't use the default
// lipgloss.NewStyle function.
// That function will use the color profile from the os.Stdin, which is the
// server, not the client.
// We provide a MakeRenderer function in the bubbletea middleware package,
// so you can easily get the correct renderer for the current session, and
// use it to create the styles.
// The recommended way to use these styles is to then pass them down to
// your Bubble Tea model.
renderer := bubbletea.MakeRenderer(s)
txtStyle := renderer.NewStyle().Foreground(lipgloss.Color("10"))
quitStyle := renderer.NewStyle().Foreground(lipgloss.Color("8"))
m := model{
term: pty.Term,
width: pty.Window.Width,
height: pty.Window.Height,
txtStyle: txtStyle,
quitStyle: quitStyle,
}
return m, []tea.ProgramOption{tea.WithAltScreen()}
}
// Just a generic tea.Model to demo terminal information of ssh.
type model struct {
term string
width int
height int
txtStyle lipgloss.Style
quitStyle lipgloss.Style
}
func (m model) Init() tea.Cmd {
return nil
}
func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
switch msg := msg.(type) {
case tea.WindowSizeMsg:
m.height = msg.Height
m.width = msg.Width
case tea.KeyMsg:
switch msg.String() {
case "q", "ctrl+c":
return m, tea.Quit
}
}
return m, nil
}
func (m model) View() string {
s := fmt.Sprintf("Your term is %s\nYour window size is %dx%d", m.term, m.width, m.height)
return m.txtStyle.Render(s) + "\n\n" + m.quitStyle.Render("Press 'q' to quit\n")
}
And when you hit that port with ssh on an instance with a Nano unikernel you get this and pressing q works!
Your term is xterm-256color
Your window size is 180x51
Press 'q' to quit