libvirt run: Add --ssh-wait flag for test integration

Add a new `--ssh-wait` flag to `bcvk libvirt run` that waits for SSH
to become available without entering an interactive shell. This is
designed for use in test suites that need to verify a VM is fully
booted and SSH-ready before proceeding with additional operations.

The `--ssh` flag behavior now also starts using this code,
as does some of the integration tests.

Assisted-by: Claude Code (Sonnet 4.5)
Signed-off-by: Colin Walters <walters@verbum.org>

Author: cgwalters

crates/integration-tests/src/tests/libvirt_port_forward.rs

@@ -272,6 +272,7 @@ fn test_libvirt_port_forward_connectivity() -> Result<()> {
         &format!("{}:8080", host_port),
         "--filesystem",
         "ext4",
+        "--ssh-wait",
         &test_image,
     ])
     .expect("Failed to run libvirt run with port forwarding");
@@ -289,12 +290,8 @@ fn test_libvirt_port_forward_connectivity() -> Result<()> {
 
     println!("Successfully created domain: {}", domain_name);
 
-    // Wait for VM to boot and SSH to become available
-    println!("Waiting for VM to boot and SSH to become available...");
-    if let Err(e) = wait_for_ssh_available(&domain_name, 180) {
-        cleanup_domain(&domain_name);
-        panic!("Failed to establish SSH connection: {}", e);
-    }
+    // SSH is already available due to --ssh-wait flag
+    println!("✓ SSH is ready (via --ssh-wait)");
 
     // Start a simple HTTP server on port 8080 inside the VM using Python
     println!("Starting HTTP server on port 8080 inside VM...");
@@ -445,56 +442,6 @@ fn cleanup_domain(domain_name: &str) {
     }
 }
 
-/// Wait for SSH to become available on a domain with a timeout
-fn wait_for_ssh_available(
-    domain_name: &str,
-    timeout_secs: u64,
-) -> Result<(), Box<dyn std::error::Error>> {
-    let start_time = std::time::Instant::now();
-    let timeout_duration = std::time::Duration::from_secs(timeout_secs);
-
-    println!(
-        "Waiting for SSH to become available on domain: {}",
-        domain_name
-    );
-
-    loop {
-        // Check if we've exceeded the timeout before attempting SSH
-        if start_time.elapsed() >= timeout_duration {
-            return Err(format!("Timeout waiting for SSH after {} seconds", timeout_secs).into());
-        }
-
-        // Try a simple SSH command to test connectivity with a short timeout (5 seconds)
-        // This prevents each SSH attempt from hanging for the default 30 seconds
-        let ssh_test = run_bcvk(&[
-            "libvirt",
-            "ssh",
-            "--timeout",
-            "5",
-            domain_name,
-            "--",
-            "echo",
-            "ssh-ready",
-        ]);
-
-        match ssh_test {
-            Ok(output) if output.success() => {
-                println!("✓ SSH is now available");
-                return Ok(());
-            }
-            Ok(_) => {
-                // SSH command failed, but that's expected while VM is booting
-            }
-            Err(e) => {
-                println!("SSH test error (expected while booting): {}", e);
-            }
-        }
-
-        // Wait 2 seconds before next attempt (since we already waited 5 seconds for SSH timeout)
-        std::thread::sleep(std::time::Duration::from_secs(2));
-    }
-}
-
 /// Find an available port on the host
 fn find_available_port() -> Result<u16> {
     use std::net::TcpListener;

crates/integration-tests/src/tests/libvirt_verb.rs

@@ -486,46 +486,6 @@ fn check_libvirt_supports_readonly_virtiofs() -> Result<bool> {
     Ok(supports_readonly)
 }
 
-/// Wait for SSH to become available on a domain with a timeout
-fn wait_for_ssh_available(
-    domain_name: &str,
-    timeout_secs: u64,
-) -> Result<(), Box<dyn std::error::Error>> {
-    let start_time = std::time::Instant::now();
-    let timeout_duration = std::time::Duration::from_secs(timeout_secs);
-
-    println!(
-        "Waiting for SSH to become available on domain: {}",
-        domain_name
-    );
-
-    loop {
-        // Try a simple SSH command to test connectivity
-        let ssh_test = run_bcvk(&["libvirt", "ssh", domain_name, "--", "echo", "ssh-ready"]);
-
-        match ssh_test {
-            Ok(output) if output.success() => {
-                println!("✓ SSH is now available");
-                return Ok(());
-            }
-            Ok(_) => {
-                // SSH command failed, but that's expected while VM is booting
-            }
-            Err(e) => {
-                println!("SSH test error (expected while booting): {}", e);
-            }
-        }
-
-        // Check if we've exceeded the timeout
-        if start_time.elapsed() >= timeout_duration {
-            return Err(format!("Timeout waiting for SSH after {} seconds", timeout_secs).into());
-        }
-
-        // Wait 5 seconds before next attempt
-        std::thread::sleep(std::time::Duration::from_secs(5));
-    }
-}
-
 /// Test VM startup and shutdown with libvirt run
 fn test_libvirt_run_vm_lifecycle() -> Result<()> {
     let bck = get_bck_command()?;
@@ -664,6 +624,7 @@ fn test_libvirt_run_bind_storage_ro() -> Result<()> {
         "--bind-storage-ro",
         "--filesystem",
         "ext4",
+        "--ssh-wait",
         &test_image,
     ])
     .expect("Failed to run libvirt run with --bind-storage-ro");
@@ -730,13 +691,10 @@ fn test_libvirt_run_bind_storage_ro() -> Result<()> {
     println!("✓ Container storage mount is configured as read-only");
     println!("✓ hoststorage tag is present in filesystem configuration");
 
-    // Wait for VM to boot and SSH to become available
-    if let Err(e) = wait_for_ssh_available(&domain_name, 180) {
-        cleanup_domain(&domain_name);
-        panic!("Failed to establish SSH connection: {}", e);
-    }
+    // SSH is already available due to --ssh-wait flag
+    println!("✓ SSH is ready (via --ssh-wait)");
 
-    // Wait for VM to boot and automatic mount to complete
+    // Wait for automatic mount to complete
     println!("Waiting for VM to boot and automatic mount to complete...");
     std::thread::sleep(std::time::Duration::from_secs(10));
 

crates/kit/src/libvirt/run.rs

@@ -19,6 +19,9 @@ use crate::libvirt::domain::VirtiofsFilesystem;
 use crate::utils::parse_memory_to_mb;
 use crate::xml_utils;
 
+/// SSH wait timeout in seconds
+const SSH_WAIT_TIMEOUT_SECONDS: u64 = 180;
+
 /// Create a virsh command with optional connection URI
 pub(super) fn virsh_command(connect_uri: Option<&str>) -> Result<std::process::Command> {
     let mut cmd = std::process::Command::new("virsh");
@@ -253,6 +256,10 @@ pub struct LibvirtRunOpts {
     #[clap(long)]
     pub ssh: bool,
 
+    /// Wait for SSH to become available and verify connectivity (for testing)
+    #[clap(long, conflicts_with = "ssh")]
+    pub ssh_wait: bool,
+
     /// Mount host container storage (RO) at /run/host-container-storage
     #[clap(long = "bind-storage-ro")]
     pub bind_storage_ro: bool,
@@ -319,6 +326,60 @@ impl LibvirtRunOpts {
     }
 }
 
+/// Wait for SSH to become available on a libvirt domain
+///
+/// Polls SSH connectivity by attempting simple commands until successful or timeout.
+fn wait_for_ssh_ready(
+    global_opts: &crate::libvirt::LibvirtOptions,
+    domain_name: &str,
+    timeout_secs: u64,
+) -> Result<()> {
+    use std::time::Duration;
+
+    debug!(
+        "Waiting for SSH to become available on domain '{}' (timeout: {}s)",
+        domain_name, timeout_secs
+    );
+
+    // Create progress bar
+    let pb = crate::boot_progress::create_boot_progress_bar();
+    pb.set_message("Waiting for SSH to become available...");
+
+    // Clone values for closure
+    let global_opts_clone = global_opts.clone();
+    let domain_name_clone = domain_name.to_string();
+
+    // Use shared polling function with libvirt-specific test
+    let (_elapsed, pb) = crate::utils::wait_for_readiness(
+        pb,
+        "Waiting for SSH",
+        || {
+            // Create a test SSH connection with short timeout
+            let ssh_opts = crate::libvirt::ssh::LibvirtSshOpts {
+                domain_name: domain_name_clone.clone(),
+                user: "root".to_string(),
+                command: vec!["true".to_string()], // Simple command to test connectivity
+                strict_host_keys: false,
+                timeout: 5, // Short timeout for each attempt
+                log_level: "ERROR".to_string(),
+                extra_options: vec![],
+                suppress_output: true, // Suppress error messages during connectivity testing
+            };
+
+            // Try to connect
+            match crate::libvirt::ssh::run_ssh_impl(&global_opts_clone, ssh_opts) {
+                Ok(_) => Ok(true),
+                Err(_) => Ok(false),
+            }
+        },
+        Duration::from_secs(timeout_secs),
+        Duration::from_secs(2), // Poll every 2 seconds
+    )?;
+
+    pb.finish_and_clear();
+    Ok(())
+}
+
 /// Execute the libvirt run command
 pub fn run(global_opts: &crate::libvirt::LibvirtOptions, opts: LibvirtRunOpts) -> Result<()> {
     use crate::images;
@@ -445,12 +506,21 @@ pub fn run(global_opts: &crate::libvirt::LibvirtOptions, opts: LibvirtRunOpts) -
         }
     }
 
-    if opts.ssh {
+    if opts.ssh_wait {
+        // Wait for SSH to be ready and verify connectivity
+        wait_for_ssh_ready(global_opts, &vm_name, SSH_WAIT_TIMEOUT_SECONDS)?;
+        println!("Ready; use bcvk libvirt ssh to connect");
+        Ok(())
+    } else if opts.ssh {
+        // Wait for SSH then enter interactive shell
+        wait_for_ssh_ready(global_opts, &vm_name, SSH_WAIT_TIMEOUT_SECONDS)?;
+
         // Use the libvirt SSH functionality directly
         let ssh_opts = crate::libvirt::ssh::LibvirtSshOpts {
             domain_name: vm_name,
             user: "root".to_string(),
             command: vec![],
+            suppress_output: false,
             strict_host_keys: false,
             timeout: 30,
             log_level: "ERROR".to_string(),

crates/kit/src/libvirt/ssh.rs

@@ -46,6 +46,10 @@ pub struct LibvirtSshOpts {
     /// Extra SSH options in key=value format
     #[clap(long)]
     pub extra_options: Vec<String>,
+
+    /// Suppress stdout/stderr output (for connectivity testing)
+    #[clap(skip)]
+    pub suppress_output: bool,
 }
 
 /// SSH configuration extracted from domain metadata
@@ -316,13 +320,17 @@ impl LibvirtSshOpts {
                 .map_err(|e| eyre!("Failed to execute SSH command: {}", e))?;
 
             if !output.stdout.is_empty() {
-                // Forward stdout to parent process
-                print!("{}", String::from_utf8_lossy(&output.stdout));
+                if !self.suppress_output {
+                    // Forward stdout to parent process
+                    print!("{}", String::from_utf8_lossy(&output.stdout));
+                }
                 debug!("SSH stdout: {}", String::from_utf8_lossy(&output.stdout));
             }
             if !output.stderr.is_empty() {
-                // Forward stderr to parent process
-                eprint!("{}", String::from_utf8_lossy(&output.stderr));
+                if !self.suppress_output {
+                    // Forward stderr to parent process
+                    eprint!("{}", String::from_utf8_lossy(&output.stderr));
+                }
                 debug!("SSH stderr: {}", String::from_utf8_lossy(&output.stderr));
             }
 

crates/kit/src/libvirt/start.rs

@@ -45,6 +45,7 @@ pub fn run(global_opts: &crate::libvirt::LibvirtOptions, opts: LibvirtStartOpts)
                 timeout: 30,
                 log_level: "ERROR".to_string(),
                 extra_options: vec![],
+                suppress_output: false,
             };
             return crate::libvirt::ssh::run(global_opts, ssh_opts);
         }
@@ -81,6 +82,7 @@ pub fn run(global_opts: &crate::libvirt::LibvirtOptions, opts: LibvirtStartOpts)
             timeout: 30,
             log_level: "ERROR".to_string(),
             extra_options: vec![],
+            suppress_output: false,
         };
         crate::libvirt::ssh::run(global_opts, ssh_opts)
     } else {

crates/kit/src/run_ephemeral_ssh.rs

@@ -3,8 +3,7 @@ use color_eyre::Result;
 use indicatif::ProgressBar;
 use std::os::unix::process::CommandExt;
 use std::process::{Command, Stdio};
-use std::thread;
-use std::time::{Duration, Instant};
+use std::time::Duration;
 use tracing::debug;
 
 use crate::run_ephemeral::{run_detached, RunEphemeralOpts};
@@ -116,38 +115,32 @@ pub fn wait_for_ssh_ready(
     let timeout = timeout.unwrap_or(SSH_TIMEOUT);
     let (_, progress) = wait_for_vm_ssh(container_name, Some(timeout), progress)?;
 
-    debug!("Polling SSH connectivity...",);
-    let start_time = Instant::now();
+    debug!("Polling SSH connectivity...");
 
     // Use SSH options optimized for connectivity testing
     let ssh_options = crate::ssh::SshConnectionOptions::for_connectivity_test();
-
-    let mut attempt = 0;
-    while start_time.elapsed() < timeout {
-        attempt += 1;
-        progress.set_message(format!("Polling for SSH readiness (attempt {attempt})"));
-
-        // Try to connect via SSH and run a simple command using the centralized SSH function
-        let status = crate::ssh::connect(
-            container_name,
-            vec!["true".to_string()], // Just run 'true' to test connectivity
-            &ssh_options,
-        );
-
-        if let Ok(exit_status) = status {
-            if exit_status.success() {
-                debug!("SSH connection successful, VM is ready");
-                return Ok((start_time.elapsed(), progress));
+    let container_name = container_name.to_string();
+
+    // Use shared polling function with container-specific test
+    crate::utils::wait_for_readiness(
+        progress,
+        "Waiting for SSH",
+        || {
+            // Try to connect via SSH and run a simple command
+            let status = crate::ssh::connect(
+                &container_name,
+                vec!["true".to_string()], // Just run 'true' to test connectivity
+                &ssh_options,
+            );
+
+            match status {
+                Ok(exit_status) if exit_status.success() => Ok(true),
+                _ => Ok(false),
             }
-        }
-
-        thread::sleep(Duration::from_secs(1));
-    }
-
-    Err(color_eyre::eyre::eyre!(
-        "Timeout waiting for SSH connectivity after {}s",
-        timeout.as_secs()
-    ))
+        },
+        timeout,
+        Duration::from_secs(1), // Poll every 1 second
+    )
 }
 
 /// Run an ephemeral pod and immediately SSH into it, with lifecycle binding