Serve concurrent terminal sessions with exact teardown

This commit is contained in:
ms2k 2026-07-18 02:51:19 +09:00
commit 254bb37764
5 changed files with 278 additions and 51 deletions

View file

@ -20,8 +20,9 @@ use lazaret_storage_catalog::Catalog;
use crate::control::ControlSessionAdmission;
use crate::control::{
ControlError, InteractiveSessionExecutor, NonInteractiveSessionExecutor, PublishedControlFace,
SessionRegistry, StagedControlFace, TrustAnchor,
AcceptedControlConnection, ControlError, InteractiveSessionExecutor,
NonInteractiveSessionExecutor, PublishedControlFace, SessionRegistry, StagedControlFace,
TrustAnchor,
};
use crate::domain_store::{DomainStoreMutation, HostCoordinatorDomainStore};
use crate::git_routing::GitRoutingCoordinator;
@ -682,7 +683,17 @@ pub fn serve_production_control(
.adapter()
.published_control()
.ok_or(ControlError::StaleGeneration)?;
face.serve_next_coordinated(
let connection = face.accept_connection()?;
serve_production_control_connection(lifecycle, &face, connection)
}
pub(crate) fn serve_production_control_connection(
lifecycle: &Mutex<ProductionHostCoordinatorLifecycle<'_, '_>>,
face: &PublishedControlFace,
connection: AcceptedControlConnection,
) -> Result<(), ControlError> {
face.serve_connection_coordinated(
connection,
|admission| {
lifecycle
.lock()

View file

@ -20,13 +20,15 @@ use lazaret_linux::filesystem::try_exclusive_flock;
use lazaret_runtime_supervisor::sandbox::production_sandbox_layout;
use rustix::fs::{Mode, OFlags, openat};
use crate::bootstrap::{ProductionHostCoordinatorLifecycle, serve_production_control};
use crate::bootstrap::{ProductionHostCoordinatorLifecycle, serve_production_control_connection};
use crate::control::ControlError;
use crate::domain_store::HostCoordinatorDomainStore;
use crate::production::{BootstrapOwnerGuard, ProductionColdStart, ProductionColdStartError};
use crate::production_selector::{InstalledProductPaths, select_production_cold_start};
use crate::waiter::BootstrapWaiterFace;
pub const PRIVATE_CLIENT_BOOTSTRAP_ROLE: &str = "--lazaret-private-client-bootstrap-v1";
const MAX_CONCURRENT_CONNECTIONS: usize = 32;
#[derive(Debug)]
pub struct PrivateClientBootstrapError(String);
@ -169,47 +171,108 @@ fn serve_published(
.ok_or_else(|| ProductionColdStartError::service("control face is absent"))?
.set_nonblocking(true)
.map_err(ProductionColdStartError::service)?;
loop {
{
let mut lifecycle = lifecycle.lock().map_err(|_| {
ProductionColdStartError::service("production lifecycle lock is poisoned")
})?;
lifecycle
.reconcile_published_bootstrap_owner(
bootstrap_owner_session_id,
bootstrap_owner
.is_live()
.map_err(ProductionColdStartError::service)?,
)
.map_err(ProductionColdStartError::service)?;
if lifecycle.adapter().published_control().is_none() {
return Ok(());
std::thread::scope(|scope| {
let (completed, completions) =
std::sync::mpsc::channel::<(u64, Result<(), ControlError>)>();
let mut active: BTreeMap<u64, std::os::unix::net::UnixStream> = BTreeMap::new();
let mut next_connection = 0_u64;
loop {
while let Ok((connection_id, result)) = completions.try_recv() {
active.remove(&connection_id);
connection_completion(result, false)?;
}
waiters
.borrow_mut()
.reconcile_published(&mut lifecycle)
.map_err(ProductionColdStartError::service)?;
}
match serve_production_control(lifecycle) {
Ok(()) => {}
Err(crate::control::ControlError::Io(error))
if error.kind() == std::io::ErrorKind::WouldBlock =>
{
std::thread::sleep(std::time::Duration::from_millis(10));
let face = {
let mut lifecycle = lifecycle.lock().map_err(|_| {
ProductionColdStartError::service("production lifecycle lock is poisoned")
})?;
let Some(face) = lifecycle.adapter().published_control() else {
for socket in active.values() {
let _ = socket.shutdown(std::net::Shutdown::Both);
}
drop(lifecycle);
while !active.is_empty() {
wait_for_connection_completion(&completions, &mut active, true)?;
}
return Ok(());
};
lifecycle
.reconcile_published_bootstrap_owner(
bootstrap_owner_session_id,
bootstrap_owner
.is_live()
.map_err(ProductionColdStartError::service)?,
)
.map_err(ProductionColdStartError::service)?;
waiters
.borrow_mut()
.reconcile_published(&mut lifecycle)
.map_err(ProductionColdStartError::service)?;
face
};
if active.len() == MAX_CONCURRENT_CONNECTIONS {
wait_for_connection_completion(&completions, &mut active, false)?;
continue;
}
match face.accept_connection() {
Ok(connection) => {
let cancellation = connection
.cancellation_handle()
.map_err(ProductionColdStartError::service)?;
let connection_id = next_connection;
next_connection = next_connection.checked_add(1).ok_or_else(|| {
ProductionColdStartError::service("control connection counter overflow")
})?;
active.insert(connection_id, cancellation);
let completed = completed.clone();
scope.spawn(move || {
let result =
serve_production_control_connection(lifecycle, &face, connection);
let _ = completed.send((connection_id, result));
});
}
Err(crate::control::ControlError::Io(error))
if error.kind() == std::io::ErrorKind::WouldBlock =>
{
std::thread::sleep(std::time::Duration::from_millis(10));
}
Err(error) => return Err(ProductionColdStartError::service(error)),
}
Err(error) => return Err(ProductionColdStartError::service(error)),
}
if lifecycle
.lock()
.map_err(|_| {
ProductionColdStartError::service("production lifecycle lock is poisoned")
})?
.adapter()
.published_control()
.is_none()
})
}
fn wait_for_connection_completion(
completions: &std::sync::mpsc::Receiver<(u64, Result<(), ControlError>)>,
active: &mut BTreeMap<u64, std::os::unix::net::UnixStream>,
cancellation_requested: bool,
) -> Result<(), ProductionColdStartError> {
let (connection_id, result) = completions
.recv()
.map_err(ProductionColdStartError::service)?;
active.remove(&connection_id);
connection_completion(result, cancellation_requested)
}
fn connection_completion(
result: Result<(), ControlError>,
cancellation_requested: bool,
) -> Result<(), ProductionColdStartError> {
match result {
Ok(()) => Ok(()),
Err(ControlError::Io(error))
if cancellation_requested
&& matches!(
error.kind(),
std::io::ErrorKind::UnexpectedEof
| std::io::ErrorKind::ConnectionReset
| std::io::ErrorKind::ConnectionAborted
| std::io::ErrorKind::BrokenPipe
| std::io::ErrorKind::NotConnected
) =>
{
return Ok(());
Ok(())
}
Err(error) => Err(ProductionColdStartError::service(error)),
}
}
@ -463,4 +526,32 @@ mod tests {
.map(OsString::from);
assert!(parse_arguments(&extra).is_err());
}
#[test]
fn withdrawal_suppresses_only_cancelled_socket_io() {
assert!(
connection_completion(
Err(ControlError::Io(std::io::Error::new(
std::io::ErrorKind::UnexpectedEof,
"withdrawn socket",
))),
true,
)
.is_ok()
);
assert!(
connection_completion(
Err(ControlError::Io(std::io::Error::new(
std::io::ErrorKind::UnexpectedEof,
"ordinary socket failure",
))),
false,
)
.is_err()
);
assert!(
connection_completion(Err(ControlError::Lifecycle("teardown failed".into())), true,)
.is_err()
);
}
}

View file

@ -838,6 +838,16 @@ pub struct PublishedControlFace {
interactive_executor: Arc<dyn InteractiveSessionExecutor>,
}
pub(crate) struct AcceptedControlConnection {
socket: std::os::unix::net::UnixStream,
}
impl AcceptedControlConnection {
pub(crate) fn cancellation_handle(&self) -> std::io::Result<std::os::unix::net::UnixStream> {
self.socket.try_clone()
}
}
pub trait NonInteractiveSessionExecutor: Send + Sync {
/// # Errors
/// Returns a runtime, stream, or exact-session execution failure.
@ -1514,6 +1524,25 @@ impl PublishedControlFace {
/// failure.
pub fn serve_next_coordinated<Admit, Disconnect>(
&self,
admit: Admit,
disconnect: Disconnect,
) -> Result<(), ControlError>
where
Admit: FnMut(ControlSessionAdmission) -> Result<(), ControlError>,
Disconnect: FnMut(SessionId) -> Result<(), ControlError>,
{
let connection = self.accept_connection()?;
self.serve_connection_coordinated(connection, admit, disconnect)
}
pub(crate) fn accept_connection(&self) -> Result<AcceptedControlConnection, ControlError> {
let (socket, _) = self.listener.accept()?;
Ok(AcceptedControlConnection { socket })
}
pub(crate) fn serve_connection_coordinated<Admit, Disconnect>(
&self,
connection: AcceptedControlConnection,
mut admit: Admit,
mut disconnect: Disconnect,
) -> Result<(), ControlError>
@ -1521,7 +1550,7 @@ impl PublishedControlFace {
Admit: FnMut(ControlSessionAdmission) -> Result<(), ControlError>,
Disconnect: FnMut(SessionId) -> Result<(), ControlError>,
{
let (socket, _) = self.listener.accept()?;
let socket = connection.socket;
let peer = socket_peercred(&socket).map_err(io::Error::from)?;
require_peer_uid(peer.uid.as_raw(), self.uid)?;
self.validate_publication()?;

View file

@ -1713,15 +1713,18 @@ impl SessionTerminalProcess for RunscSessionTerminal {
}
fn reap_exec(&mut self) -> io::Result<ProcessExit> {
if let Some(exit) = self.exit {
return Ok(exit);
}
let exit = child_broker().reap_status(
self.exec
.take()
.ok_or_else(|| io::Error::other("exec client exit is unavailable"))?,
)?;
self.exit = Some(exit);
let exit = if let Some(exit) = self.exit {
exit
} else {
let exit = child_broker().reap_status(
self.exec
.take()
.ok_or_else(|| io::Error::other("exec client exit is unavailable"))?,
)?;
self.exit = Some(exit);
exit
};
remove_pid_file(&self.request.internal_pid_file)?;
Ok(exit)
}

View file

@ -302,6 +302,23 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
.store
.apply(&[DomainStoreMutation::PutSession(session.clone())])
.map_err(failure)?;
let second_session_id = SessionId::new([41; 32]);
let mut second_session = Session {
generation_id: GenerationId::new(7),
session_id: second_session_id,
cli_id: ClientId::new([42; 32]),
session_anchor_id: SessionAnchorId::new([43; 32]),
session_phase: SessionPhase::Registering,
};
fixture
.store
.apply(&[DomainStoreMutation::PutSession(second_session.clone())])
.map_err(failure)?;
second_session.session_phase = SessionPhase::Attached;
fixture
.store
.apply(&[DomainStoreMutation::PutSession(second_session)])
.map_err(failure)?;
session.session_phase = SessionPhase::Attached;
fixture
.store
@ -317,6 +334,9 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
let count = workspace.join("session-count");
let finish = workspace.join("session-finish");
let signals = workspace.join("session-signals");
let second_count = workspace.join("session-second-count");
let second_finish = workspace.join("session-second-finish");
let second_signals = workspace.join("session-second-signals");
let line = "x".repeat(1024);
let script = format!(
"trap 'echo INT >> /workspace/session-signals' INT; \
@ -326,6 +346,15 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
n=0; while test ! -e /workspace/session-finish; do \
echo '{line}'; n=$((n+1)); echo $n > /workspace/session-count; done; exit 23"
);
let second_line = "y".repeat(1024);
let second_script = format!(
"trap 'echo INT >> /workspace/session-second-signals' INT; \
trap 'echo QUIT >> /workspace/session-second-signals' QUIT; \
trap 'echo TSTP >> /workspace/session-second-signals' TSTP; \
trap 'echo WINCH >> /workspace/session-second-signals' WINCH; \
n=0; while test ! -e /workspace/session-second-finish; do \
echo '{second_line}'; n=$((n+1)); echo $n > /workspace/session-second-count; done; exit 42"
);
let request = TerminalAcquisition {
domain_id: fixture.configuration.domain_id(),
generation_id: GenerationId::new(7),
@ -344,7 +373,20 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
Arc::new(NoActivePrompt),
);
engine.acquire(&request).map_err(failure)?;
engine
.acquire(&TerminalAcquisition {
session_id: second_session_id,
internal_pid_file: runtime_domain.join("session-second-internal.pid"),
argv: vec![
b"/bin/sh".to_vec(),
b"-c".to_vec(),
second_script.into_bytes(),
],
..request.clone()
})
.map_err(failure)?;
wait_for_path(&count)?;
wait_for_path(&second_count)?;
let running = fixture
.store
.snapshot()
@ -384,6 +426,9 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
.map_err(failure)?;
}
wait_for_signal_lines(&mut engine, session_id, &signals, 4)?;
if second_signals.exists() {
return Err(failure("terminal signals crossed Session IDs"));
}
let mut forwarded_before_prompt = 0_usize;
for _ in 0..2_000 {
let messages = engine
@ -417,11 +462,21 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
.map_err(failure)?;
thread::sleep(Duration::from_millis(300));
let before = fs::read_to_string(&count)?;
thread::sleep(Duration::from_millis(300));
let second_before = fs::read_to_string(&second_count)?;
for _ in 0..60 {
engine
.poll(second_session_id, GenerationId::new(7))
.map_err(failure)?;
thread::sleep(Duration::from_millis(5));
}
let after = fs::read_to_string(&count)?;
let second_after = fs::read_to_string(&second_count)?;
if before != after {
return Err(failure("prompt route did not apply PTY backpressure"));
}
if second_before == second_after {
return Err(failure("one Prompt paused another Session Terminal"));
}
session.session_phase = SessionPhase::Attached;
fixture
.store
@ -466,11 +521,49 @@ fn run_session_terminal_matrix() -> Result<(), MatrixError> {
}
if exit.and_then(lazaret_core::records::SessionExit::exit_code) != Some(23)
|| engine.contains(session_id)
|| !engine.contains(second_session_id)
{
return Err(failure(
"Session Terminal exit or no-reattachment cleanup failed",
));
}
let second_before_first_exit = fs::read_to_string(&second_count)?;
for _ in 0..200 {
engine
.poll(second_session_id, GenerationId::new(7))
.map_err(failure)?;
if fs::read_to_string(&second_count).ok().as_deref()
!= Some(second_before_first_exit.as_str())
{
break;
}
thread::sleep(Duration::from_millis(5));
}
if fs::read_to_string(&second_count).ok().as_deref() == Some(second_before_first_exit.as_str())
{
return Err(failure("first Session exit paused the second Session"));
}
fs::write(&second_finish, b"finish")?;
let mut second_exit = None;
for _ in 0..2_000 {
for message in engine
.poll(second_session_id, GenerationId::new(7))
.map_err(failure)?
{
if let SessionTerminalMessage::Exit(outcome) = message {
second_exit = Some(outcome);
}
}
if second_exit.is_some() {
break;
}
thread::sleep(Duration::from_millis(5));
}
if second_exit.and_then(lazaret_core::records::SessionExit::exit_code) != Some(42)
|| engine.contains(second_session_id)
{
return Err(failure("second Session exit routing or cleanup failed"));
}
runtime
.stop(&SandboxCoordinator::new(&fixture.store))
.map_err(failure)