WEFT_OS/crates/weft-appd/src/runtime.rs

827 lines
30 KiB
Rust
Raw Normal View History

use std::path::PathBuf;
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
use std::time::Duration;
use tokio::io::{AsyncBufReadExt, AsyncWriteExt, BufReader, BufWriter};
use weft_ipc_types::AppdToCompositor;
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
use crate::Registry;
use crate::compositor_client::CompositorSender;
use crate::ipc::{AppStateKind, Response};
use crate::launch::LaunchPackage;
use crate::mount::Mount;
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
/// The per-session relay between the component's IPC socket and the app
/// bridge. It belongs to the session: closing it stops the relay task and
/// removes the socket, whether or not the component ever connected.
pub(crate) struct IpcRelay {
task: tokio::task::JoinHandle<()>,
socket_path: PathBuf,
}
impl IpcRelay {
/// Completes when the relay stops: the component closed its connection
/// or the listener failed.
async fn ended(&mut self) {
let _ = (&mut self.task).await;
}
}
impl Drop for IpcRelay {
fn drop(&mut self) {
self.task.abort();
let _ = std::fs::remove_file(&self.socket_path);
}
}
/// Listens on `socket_path` for the session's component and relays
/// newline-delimited messages between it and the returned sender (towards
/// the component) and the broadcast channel (from the component).
#[cfg(unix)]
pub(crate) fn spawn_ipc_relay(
session_id: u64,
socket_path: PathBuf,
broadcast: tokio::sync::broadcast::Sender<Response>,
) -> std::io::Result<(tokio::sync::mpsc::Sender<String>, IpcRelay)> {
let _ = std::fs::remove_file(&socket_path);
let listener = tokio::net::UnixListener::bind(&socket_path)?;
{
use std::os::unix::fs::PermissionsExt;
if let Err(e) =
std::fs::set_permissions(&socket_path, std::fs::Permissions::from_mode(0o600))
{
let _ = std::fs::remove_file(&socket_path);
return Err(e);
}
}
let (html_to_wasm_tx, mut html_to_wasm_rx) = tokio::sync::mpsc::channel::<String>(64);
let task = tokio::spawn(async move {
let Ok((stream, _)) = listener.accept().await else {
tracing::warn!(session_id, "IPC relay: failed to accept connection");
return;
};
// One component per session: no further connections are accepted.
drop(listener);
tracing::debug!(session_id, "IPC relay: component connected");
let (reader, writer) = tokio::io::split(stream);
let mut reader = BufReader::new(reader);
let mut writer = BufWriter::new(writer);
loop {
let mut line = String::new();
tokio::select! {
n = reader.read_line(&mut line) => {
match n {
Ok(0) | Err(_) => break,
Ok(_) => {
let payload = line.trim_end().to_owned();
let _ = broadcast.send(Response::IpcMessage { session_id, payload });
}
}
}
msg = html_to_wasm_rx.recv() => {
match msg {
Some(payload) => {
let mut data = payload;
data.push('\n');
if writer.write_all(data.as_bytes()).await.is_err()
|| writer.flush().await.is_err()
{
break;
}
}
None => break,
}
}
}
}
});
Ok((html_to_wasm_tx, IpcRelay { task, socket_path }))
}
#[cfg(not(unix))]
pub(crate) fn spawn_ipc_relay(
_session_id: u64,
_socket_path: PathBuf,
_broadcast: tokio::sync::broadcast::Sender<Response>,
) -> std::io::Result<(tokio::sync::mpsc::Sender<String>, IpcRelay)> {
Err(std::io::ErrorKind::Unsupported.into())
}
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
const READY_TIMEOUT: Duration = Duration::from_secs(30);
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
/// Environment variable carrying a session's readiness token to its children.
///
/// A child reports readiness by printing `READY <token>` on its own stdout.
/// Application code shares that stdout (Wasm guests through inherited stdio,
/// pages through `window.alert`) but cannot read the token, because the
/// runtime passes the guest only the variables it names and pages cannot read
/// the host process environment.
const READY_TOKEN_ENV: &str = "WEFT_READY_TOKEN";
/// The application's close policy, `immediate` or `ask`, for the app shell.
const CLOSE_POLICY_ENV: &str = "WEFT_APP_CLOSE";
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
/// Environment variable carrying the session's application bridge credential
/// to weft-app-shell.
const BRIDGE_TOKEN_ENV: &str = "WEFT_BRIDGE_TOKEN";
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
/// Longest line read from a child's output; longer lines are split.
const MAX_LINE: u64 = 64 * 1024;
/// A random 128-bit credential as 32 hex characters.
pub(crate) fn random_token() -> std::io::Result<String> {
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
use std::io::Read;
let mut bytes = [0u8; 16];
std::fs::File::open("/dev/urandom")?.read_exact(&mut bytes)?;
Ok(bytes.iter().map(|b| format!("{b:02x}")).collect())
}
/// Reads one line of at most MAX_LINE bytes, decoded lossily. Returns None at
/// end of output.
async fn read_child_line<R: tokio::io::AsyncBufRead + Unpin>(
reader: &mut R,
) -> std::io::Result<Option<String>> {
use tokio::io::AsyncReadExt;
let mut line = Vec::new();
let n = (&mut *reader)
.take(MAX_LINE)
.read_until(b'\n', &mut line)
.await?;
if n == 0 {
return Ok(None);
}
Ok(Some(String::from_utf8_lossy(&line).trim_end().to_owned()))
}
/// `systemd-run` options that place the runtime in a transient user scope.
///
/// A scope runs the command in the foreground as systemd-run's own process,
/// so its stdout and exit status reach appd directly. `--wait` only applies
/// to service units; systemd rejects it together with `--scope`.
const SYSTEMD_SCOPE_ARGS: &[&str] = &[
"--user",
"--scope",
"--collect",
"--slice=weft-apps.slice",
"-p",
"CPUQuota=200%",
"-p",
"MemoryMax=512M",
];
fn systemd_cgroup_available() -> bool {
if std::env::var("WEFT_DISABLE_CGROUP").is_ok() {
return false;
}
let Ok(runtime_dir) = std::env::var("XDG_RUNTIME_DIR") else {
return false;
};
std::path::Path::new(&runtime_dir)
.join("systemd/private")
.exists()
}
async fn kill_portal(portal: Option<(PathBuf, tokio::process::Child)>) {
if let Some((sock, mut child)) = portal {
let _ = child.kill().await;
let _ = child.wait().await;
let _ = std::fs::remove_file(&sock);
}
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
fn spawn_app_shell(
bin: &str,
session_id: u64,
package: &LaunchPackage,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
token: &str,
bridge_token: Option<String>,
wayland: Option<std::os::unix::net::UnixStream>,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
) -> std::io::Result<tokio::process::Child> {
let mut command = tokio::process::Command::new(bin);
match wayland {
// The app shell's only Wayland connection is the one created for its
// session, handed over as its standard input: WAYLAND_SOCKET names
// that descriptor, and no display socket is offered to connect
// anywhere else. Only this child receives the descriptor.
Some(stream) => {
command
.stdin(std::process::Stdio::from(std::os::fd::OwnedFd::from(
stream,
)))
.env("WAYLAND_SOCKET", "0")
.env_remove("WAYLAND_DISPLAY");
}
None => {
command.stdin(std::process::Stdio::null());
}
}
if let Some(bridge_token) = bridge_token {
// The page holds this credential; it must differ from the readiness
// token, which page output could otherwise use to fake readiness.
command.env(BRIDGE_TOKEN_ENV, bridge_token);
}
let child = command
.arg(&package.app_id)
.arg(session_id.to_string())
.arg("--ui")
.arg(&package.ui_entry)
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
.env(READY_TOKEN_ENV, token)
.env(CLOSE_POLICY_ENV, package.close.as_str())
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
.stdout(std::process::Stdio::piped())
.stderr(std::process::Stdio::piped())
.kill_on_drop(true)
.spawn()?;
tracing::info!(session_id, app_id = %package.app_id, bin, "app shell spawned");
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
Ok(child)
}
fn portal_socket_path(session_id: u64) -> Option<PathBuf> {
let runtime_dir = std::env::var("XDG_RUNTIME_DIR").ok()?;
let dir = PathBuf::from(runtime_dir).join("weft");
crate::private_dir(&dir).ok()?;
Some(dir.join(format!("portal-{session_id}.sock")))
}
fn spawn_file_portal(
session_id: u64,
fix(runtime): enforce effective capability grants per session Every host import was registered and callable whatever the package declared, read-only directory capabilities were preopened read-write, and appd launched packages whose capabilities it could not satisfy. The fetch import also could not be linked: the host returned a tuple where the WIT declares the response record, so any component importing weft:app/fetch failed to instantiate. One capability vocabulary now lives in weft_ipc_types::capability and is used by weft-pack, weft-appd and weft-runtime. weft-appd derives a session's grants once, before the session exists, and refuses the launch with error 403 when a capability is unknown, unsupported (hw:gpu:*) or unsatisfiable (documents without an XDG documents directory). The documents directory now comes from user-dirs.dirs instead of a guessed ~/Documents. The runtime receives grants explicitly (--grant CAP, --preopen HOST::GUEST::ro|rw), preopens read-only directories read-only, and checks the grant on every notifications, clipboard and fetch call. Fetch accepts only http and https to granted hosts, returns redirects and HTTP error statuses as responses instead of following or failing them, and bounds connect time, total time and response size. The file portal takes read-only directories with --allow-read and refuses writes there. tests/runtime/check_grants.py runs the real runtime on a probe component under each kind of grant and checks the outcome of every operation; it fails when read-only data is writable or an import ignores its grant. The appd dispatch tests that read WEFT_RUNTIME_BIN now hold the env lock shared with the tests that set it.
2026-10-09 08:08:58 +00:00
dirs: &[crate::grants::GrantedDir],
) -> Option<(PathBuf, tokio::process::Child)> {
let bin = std::env::var("WEFT_FILE_PORTAL_BIN").ok()?;
let socket = portal_socket_path(session_id)?;
let mut cmd = tokio::process::Command::new(&bin);
cmd.arg(&socket)
.stdin(std::process::Stdio::null())
.stdout(std::process::Stdio::null())
.stderr(std::process::Stdio::null());
fix(runtime): enforce effective capability grants per session Every host import was registered and callable whatever the package declared, read-only directory capabilities were preopened read-write, and appd launched packages whose capabilities it could not satisfy. The fetch import also could not be linked: the host returned a tuple where the WIT declares the response record, so any component importing weft:app/fetch failed to instantiate. One capability vocabulary now lives in weft_ipc_types::capability and is used by weft-pack, weft-appd and weft-runtime. weft-appd derives a session's grants once, before the session exists, and refuses the launch with error 403 when a capability is unknown, unsupported (hw:gpu:*) or unsatisfiable (documents without an XDG documents directory). The documents directory now comes from user-dirs.dirs instead of a guessed ~/Documents. The runtime receives grants explicitly (--grant CAP, --preopen HOST::GUEST::ro|rw), preopens read-only directories read-only, and checks the grant on every notifications, clipboard and fetch call. Fetch accepts only http and https to granted hosts, returns redirects and HTTP error statuses as responses instead of following or failing them, and bounds connect time, total time and response size. The file portal takes read-only directories with --allow-read and refuses writes there. tests/runtime/check_grants.py runs the real runtime on a probe component under each kind of grant and checks the outcome of every operation; it fails when read-only data is writable or an import ignores its grant. The appd dispatch tests that read WEFT_RUNTIME_BIN now hold the env lock shared with the tests that set it.
2026-10-09 08:08:58 +00:00
for dir in dirs {
let flag = match dir.access {
weft_ipc_types::capability::Access::Read => "--allow-read",
weft_ipc_types::capability::Access::ReadWrite => "--allow",
};
cmd.arg(flag).arg(&dir.host);
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let child = cmd.kill_on_drop(true).spawn().ok()?;
tracing::info!(session_id, socket = %socket.display(), "file portal spawned");
Some((socket, child))
}
/// A request to end a session, from `TERMINATE_APP`.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub(crate) enum Stop {
/// Close as a user close would; the application may decline.
Close,
/// Terminate without asking.
Force,
}
pub(crate) type StopSender = tokio::sync::mpsc::UnboundedSender<Stop>;
/// Ends when appd shuts down and drops the sender.
pub(crate) type StopReceiver = tokio::sync::mpsc::UnboundedReceiver<Stop>;
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
pub(crate) async fn supervise(
session_id: u64,
mut package: LaunchPackage,
fix(runtime): enforce effective capability grants per session Every host import was registered and callable whatever the package declared, read-only directory capabilities were preopened read-write, and appd launched packages whose capabilities it could not satisfy. The fetch import also could not be linked: the host returned a tuple where the WIT declares the response record, so any component importing weft:app/fetch failed to instantiate. One capability vocabulary now lives in weft_ipc_types::capability and is used by weft-pack, weft-appd and weft-runtime. weft-appd derives a session's grants once, before the session exists, and refuses the launch with error 403 when a capability is unknown, unsupported (hw:gpu:*) or unsatisfiable (documents without an XDG documents directory). The documents directory now comes from user-dirs.dirs instead of a guessed ~/Documents. The runtime receives grants explicitly (--grant CAP, --preopen HOST::GUEST::ro|rw), preopens read-only directories read-only, and checks the grant on every notifications, clipboard and fetch call. Fetch accepts only http and https to granted hosts, returns redirects and HTTP error statuses as responses instead of following or failing them, and bounds connect time, total time and response size. The file portal takes read-only directories with --allow-read and refuses writes there. tests/runtime/check_grants.py runs the real runtime on a probe component under each kind of grant and checks the outcome of every operation; it fails when read-only data is writable or an import ignores its grant. The appd dispatch tests that read WEFT_RUNTIME_BIN now hold the env lock shared with the tests that set it.
2026-10-09 08:08:58 +00:00
grants: crate::grants::SessionGrants,
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
registry: Registry,
abort_rx: StopReceiver,
compositor_tx: Option<CompositorSender>,
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
) -> anyhow::Result<()> {
let mut abort_rx = abort_rx;
let app_id = package.app_id.clone();
let app_id = app_id.as_str();
let image = package.image.take();
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
let bin = match std::env::var("WEFT_RUNTIME_BIN") {
Ok(b) => b,
Err(_) => {
tracing::debug!(session_id, %app_id, "WEFT_RUNTIME_BIN not set; skipping process spawn");
return stop_unstarted(&registry, session_id, image).await;
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
}
};
let Ok(shell_bin) = std::env::var("WEFT_APP_SHELL_BIN") else {
tracing::warn!(session_id, %app_id, "WEFT_APP_SHELL_BIN not set; no UI host to start");
return stop_unstarted(&registry, session_id, image).await;
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
};
let token = match random_token() {
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
Ok(token) => token,
Err(e) => {
tracing::error!(session_id, %app_id, error = %e, "cannot create readiness token");
return stop_unstarted(&registry, session_id, image).await;
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
}
};
let Some(ipc_socket_path) = crate::session_ipc_socket_path(session_id) else {
tracing::warn!(session_id, %app_id, "no runtime directory for the IPC socket");
return stop_unstarted(&registry, session_id, image).await;
};
let broadcast = registry.lock().await.broadcast().clone();
let relay = match spawn_ipc_relay(session_id, ipc_socket_path.clone(), broadcast) {
Ok((tx, relay)) => {
registry.lock().await.register_ipc_sender(session_id, tx);
relay
}
Err(e) => {
tracing::warn!(session_id, %app_id, error = %e, "cannot open the IPC socket");
return stop_unstarted(&registry, session_id, image).await;
}
};
fix(runtime): enforce effective capability grants per session Every host import was registered and callable whatever the package declared, read-only directory capabilities were preopened read-write, and appd launched packages whose capabilities it could not satisfy. The fetch import also could not be linked: the host returned a tuple where the WIT declares the response record, so any component importing weft:app/fetch failed to instantiate. One capability vocabulary now lives in weft_ipc_types::capability and is used by weft-pack, weft-appd and weft-runtime. weft-appd derives a session's grants once, before the session exists, and refuses the launch with error 403 when a capability is unknown, unsupported (hw:gpu:*) or unsatisfiable (documents without an XDG documents directory). The documents directory now comes from user-dirs.dirs instead of a guessed ~/Documents. The runtime receives grants explicitly (--grant CAP, --preopen HOST::GUEST::ro|rw), preopens read-only directories read-only, and checks the grant on every notifications, clipboard and fetch call. Fetch accepts only http and https to granted hosts, returns redirects and HTTP error statuses as responses instead of following or failing them, and bounds connect time, total time and response size. The file portal takes read-only directories with --allow-read and refuses writes there. tests/runtime/check_grants.py runs the real runtime on a probe component under each kind of grant and checks the outcome of every operation; it fails when read-only data is writable or an import ignores its grant. The appd dispatch tests that read WEFT_RUNTIME_BIN now hold the env lock shared with the tests that set it.
2026-10-09 08:08:58 +00:00
let portal = spawn_file_portal(session_id, &grants.dirs);
let mut cmd = if systemd_cgroup_available() {
let mut c = tokio::process::Command::new("systemd-run");
c.args(SYSTEMD_SCOPE_ARGS).arg("--").arg(&bin);
c
} else {
tokio::process::Command::new(&bin)
};
cmd.arg(app_id)
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
.arg(session_id.to_string())
.stdout(std::process::Stdio::piped())
.stderr(std::process::Stdio::piped())
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
.stdin(std::process::Stdio::null())
.env(READY_TOKEN_ENV, &token)
.kill_on_drop(true);
cmd.arg("--ipc-socket").arg(&ipc_socket_path);
cmd.arg("--module").arg(&package.module);
if let Some((ref sock, _)) = portal {
cmd.env("WEFT_FILE_PORTAL_SOCKET", sock);
}
fix(runtime): enforce effective capability grants per session Every host import was registered and callable whatever the package declared, read-only directory capabilities were preopened read-write, and appd launched packages whose capabilities it could not satisfy. The fetch import also could not be linked: the host returned a tuple where the WIT declares the response record, so any component importing weft:app/fetch failed to instantiate. One capability vocabulary now lives in weft_ipc_types::capability and is used by weft-pack, weft-appd and weft-runtime. weft-appd derives a session's grants once, before the session exists, and refuses the launch with error 403 when a capability is unknown, unsupported (hw:gpu:*) or unsatisfiable (documents without an XDG documents directory). The documents directory now comes from user-dirs.dirs instead of a guessed ~/Documents. The runtime receives grants explicitly (--grant CAP, --preopen HOST::GUEST::ro|rw), preopens read-only directories read-only, and checks the grant on every notifications, clipboard and fetch call. Fetch accepts only http and https to granted hosts, returns redirects and HTTP error statuses as responses instead of following or failing them, and bounds connect time, total time and response size. The file portal takes read-only directories with --allow-read and refuses writes there. tests/runtime/check_grants.py runs the real runtime on a probe component under each kind of grant and checks the outcome of every operation; it fails when read-only data is writable or an import ignores its grant. The appd dispatch tests that read WEFT_RUNTIME_BIN now hold the env lock shared with the tests that set it.
2026-10-09 08:08:58 +00:00
for dir in &grants.dirs {
cmd.arg("--preopen").arg(dir.preopen_arg());
}
for capability in &grants.imports {
cmd.arg("--grant").arg(capability.to_string());
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let mut session = OwnedSession {
session_id,
runtime: None,
app_shell: None,
portal,
relay: Some(relay),
image,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
compositor_tx,
client_attached: false,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
};
let mut runtime = match cmd.spawn() {
Ok(c) => c,
Err(e) => {
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
return session
.settle(&registry, &format!("failed to spawn runtime: {e}"))
.await;
}
};
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let runtime_stdout = runtime.stdout.take().expect("stdout piped");
tokio::spawn(drain_stderr(
runtime.stderr.take().expect("stderr piped"),
session_id,
"runtime",
));
session.runtime = Some(runtime);
// The component must initialize before the UI host starts.
let runtime_stdout = match wait_for_child_ready(runtime_stdout, &token, &mut abort_rx).await {
Ok(reader) => reader,
Err(reason) => {
return session
.settle(&registry, &format!("runtime {reason}"))
.await;
}
};
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
tokio::spawn(drain_stdout(runtime_stdout, session_id));
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
// With a compositor, the app shell connects only through a connection
// the compositor received for this session, so every surface it creates
// is bound to the session.
let wayland = match &session.compositor_tx {
None => None,
Some(tx) => match attach_client(tx, session_id, app_id) {
Ok(stream) => {
session.client_attached = true;
Some(stream)
}
Err(reason) => return session.settle(&registry, &reason).await,
},
};
let bridge_token = registry.lock().await.bridge_token(session_id);
let mut app_shell = match spawn_app_shell(
&shell_bin,
session_id,
&package,
&token,
bridge_token,
wayland,
) {
Ok(child) => child,
Err(e) => {
return session
.settle(&registry, &format!("failed to spawn app shell: {e}"))
.await;
}
};
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let shell_stdout = app_shell.stdout.take().expect("stdout piped");
tokio::spawn(drain_stderr(
app_shell.stderr.take().expect("stderr piped"),
session_id,
"app shell",
));
session.app_shell = Some(app_shell);
// Running requires the UI host to have presented the loaded document.
let shell_stdout = tokio::select! {
r = wait_for_child_ready(shell_stdout, &token, &mut abort_rx) => {
r.map_err(|reason| format!("app shell {reason}"))
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
status = session.runtime.as_mut().expect("runtime spawned").wait() => {
Err(format!("runtime exited before the app shell was ready ({status:?})"))
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
}
_ = session.relay.as_mut().expect("relay opened").ended() => {
Err("component closed its IPC connection before the session was ready".to_owned())
}
};
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let shell_stdout = match shell_stdout {
Ok(reader) => reader,
Err(reason) => return session.settle(&registry, &reason).await,
};
let (cancel_tx, mut cancelled) = tokio::sync::mpsc::channel(1);
tokio::spawn(watch_app_shell_stdout(
shell_stdout,
session_id,
token.clone(),
cancel_tx,
));
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
{
let mut reg = registry.lock().await;
reg.set_state(session_id, AppStateKind::Running);
let _ = reg.broadcast().send(Response::AppReady {
session_id,
app_id: app_id.to_owned(),
});
}
tracing::info!(session_id, %app_id, "app ready");
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
// Whether the user asked the application to close, so that a clean exit
// afterwards, also one the application makes after declining, is a
// close on request.
let mut close_asked = false;
let reason = loop {
tokio::select! {
status = session.runtime.as_mut().expect("runtime spawned").wait() => {
break format!("runtime exited ({status:?})");
}
status = session.app_shell.as_mut().expect("app shell spawned").wait() => {
break match status {
Ok(status) if close_asked && status.success() => {
format!("closed on request; app shell exited ({status:?})")
}
status => format!("app shell exited ({status:?})"),
};
}
stop = abort_rx.recv() => match stop {
Some(Stop::Force) => break "terminated by force".to_owned(),
Some(Stop::Close) => {
close_asked = true;
match session.close(session_id, &mut cancelled, &mut abort_rx).await {
Close::Settle(reason) => break reason,
// The application keeps running, as its close
// policy allows; its window comes to the front so
// the user sees why. A later request asks again.
Close::Cancelled => {
tracing::info!(session_id, "the application declined the close");
session.focus(session_id);
}
}
}
// The senders are dropped only when appd shuts down.
None => break "appd is shutting down".to_owned(),
},
_ = session.relay.as_mut().expect("relay opened").ended() => {
break "component closed its IPC connection".to_owned();
}
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
};
session.settle(&registry, &reason).await
}
/// How a close request ended.
enum Close {
/// The session ends, for this reason.
Settle(String),
/// The application declined the close and keeps running.
Cancelled,
}
/// The app shell's report that its page declined a close, followed by the
/// session's readiness token so page output cannot fake it.
const CLOSE_CANCELLED: &str = "CLOSE_CANCELLED";
/// Passes the app shell's close reports on to the supervisor and logs
/// everything else it prints.
async fn watch_app_shell_stdout(
mut reader: BufReader<tokio::process::ChildStdout>,
session_id: u64,
token: String,
cancelled: tokio::sync::mpsc::Sender<()>,
) {
while let Ok(Some(line)) = read_child_line(&mut reader).await {
// Earlier output without a trailing newline can precede the
// report, as it can precede READY; the token alone establishes it.
if line.rsplit_once(' ').is_some_and(|(head, t)| {
head.ends_with(CLOSE_CANCELLED) && crate::ws::tokens_match(&token, t)
}) {
// One pending report is enough; the supervisor drops stale ones
// before it asks again.
let _ = cancelled.try_send(());
} else {
tracing::debug!(session_id, stdout = %line, "child stdout");
}
}
}
/// How long an application may take to close after being asked, for
/// example to save, before its processes are terminated.
pub(crate) const CLOSE_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(5);
/// Creates the session's Wayland connection and queues the compositor's end
/// for the compositor, returning the app shell's end.
fn attach_client(
tx: &CompositorSender,
session_id: u64,
app_id: &str,
) -> Result<std::os::unix::net::UnixStream, String> {
if !tx.is_connected() {
return Err("weft-compositor is not connected, so the app cannot show a window".to_owned());
}
let (shell, compositor) = std::os::unix::net::UnixStream::pair()
.map_err(|e| format!("cannot create the app's Wayland connection: {e}"))?;
tx.try_send(crate::compositor_client::Outbound {
msg: AppdToCompositor::AttachClient {
session_id,
app_id: app_id.to_owned(),
},
fd: Some(compositor.into()),
})
.map_err(|_| "the compositor connection is not available".to_owned())?;
Ok(shell)
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
/// Marks a session that never started any process as stopped.
async fn stop_unstarted(
registry: &Registry,
session_id: u64,
image: Option<Mount>,
) -> anyhow::Result<()> {
release_image(image).await;
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let mut reg = registry.lock().await;
reg.set_state(session_id, AppStateKind::Stopped);
reg.remove_abort_sender(session_id);
let _ = reg.broadcast().send(Response::AppState {
session_id,
state: AppStateKind::Stopped,
});
Ok(())
}
/// The processes, IPC relay, portal, compositor association and image mount
/// a session owns. `settle` releases all of them on every exit path: spawn
/// failure, readiness failure, timeout, abort and child exit.
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
struct OwnedSession {
session_id: u64,
runtime: Option<tokio::process::Child>,
app_shell: Option<tokio::process::Child>,
portal: Option<(PathBuf, tokio::process::Child)>,
relay: Option<IpcRelay>,
/// The mounted package image, released after both children have exited.
image: Option<Mount>,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
compositor_tx: Option<CompositorSender>,
/// Whether the compositor was given this session's client connection.
client_attached: bool,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
impl OwnedSession {
/// Closes a running session as a user close would: the compositor asks
/// its windows to close and the app shell gets `CLOSE_TIMEOUT` to exit by
/// itself. Returns the terminal reason; settling terminates whatever is
/// still running.
async fn close(
&mut self,
session_id: u64,
cancelled: &mut tokio::sync::mpsc::Receiver<()>,
stops: &mut StopReceiver,
) -> Close {
// A report already read from an earlier close, for example one the
// user started in the compositor, does not answer this one. One
// still in the pipe can: the session then reads as declined while
// the app shell is still asking, and the app shell's own answer
// timeout still closes a page that does not answer.
while cancelled.try_recv().is_ok() {}
let asked = self.client_attached
&& self.compositor_tx.as_ref().is_some_and(|tx| {
tx.is_connected()
&& tx
.try_send(AppdToCompositor::AppCloseRequest { session_id }.into())
.is_ok()
});
if !asked {
return Close::Settle(
"terminate requested; the compositor could not ask the app to close".to_owned(),
);
}
let (Some(app_shell), Some(runtime), Some(relay)) = (
self.app_shell.as_mut(),
self.runtime.as_mut(),
self.relay.as_mut(),
) else {
return Close::Settle("terminate requested".to_owned());
};
let wait = async {
loop {
tokio::select! {
status = app_shell.wait() => break Close::Settle(match status {
Ok(status) if status.success() => {
format!("closed on request; app shell exited ({status:?})")
}
status => format!("app shell failed while closing ({status:?})"),
}),
status = runtime.wait() => {
break Close::Settle(format!("runtime exited while closing ({status:?})"));
}
_ = relay.ended() => break Close::Settle(
"component closed its IPC connection while closing".to_owned(),
),
Some(()) = cancelled.recv() => break Close::Cancelled,
stop = stops.recv() => match stop {
// Asking again while the application answers changes
// nothing.
Some(Stop::Close) => {}
Some(Stop::Force) => {
break Close::Settle("terminated by force while closing".to_owned());
}
None => break Close::Settle("appd is shutting down".to_owned()),
},
}
}
};
tokio::time::timeout(CLOSE_TIMEOUT, wait)
.await
.unwrap_or_else(|_| {
Close::Settle(format!(
"close requested; the app shell did not exit within {} s and was terminated",
CLOSE_TIMEOUT.as_secs()
))
})
}
/// Brings the session's window to the front with keyboard focus.
fn focus(&self, session_id: u64) {
if self.client_attached
&& let Some(tx) = &self.compositor_tx
{
let _ = tx.try_send(AppdToCompositor::AppFocusRequest { session_id }.into());
}
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
async fn settle(self, registry: &Registry, reason: &str) -> anyhow::Result<()> {
let session_id = self.session_id;
tracing::info!(session_id, reason, "stopping session");
{
let mut reg = registry.lock().await;
if !matches!(reg.state(session_id), AppStateKind::Stopping) {
reg.set_state(session_id, AppStateKind::Stopping);
let _ = reg.broadcast().send(Response::AppState {
session_id,
state: AppStateKind::Stopping,
});
}
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
kill_child(self.app_shell).await;
kill_child(self.runtime).await;
if self.client_attached
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
&& let Some(tx) = &self.compositor_tx
&& tx
.try_send(AppdToCompositor::AppSurfaceDestroyed { session_id }.into())
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
.is_err()
{
tracing::warn!(
session_id,
"compositor queue unavailable; surface not released"
);
}
release_image(self.image).await;
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
kill_portal(self.portal).await;
drop(self.relay);
let mut reg = registry.lock().await;
reg.remove_ipc_sender(session_id);
reg.set_state(session_id, AppStateKind::Stopped);
reg.remove_abort_sender(session_id);
let _ = reg.broadcast().send(Response::AppState {
session_id,
state: AppStateKind::Stopped,
});
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
Ok(())
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
}
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
/// Unmounts a package image without blocking the async workers.
async fn release_image(image: Option<Mount>) {
if let Some(image) = image {
let _ = tokio::task::spawn_blocking(move || drop(image)).await;
}
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
async fn kill_child(child: Option<tokio::process::Child>) {
if let Some(mut child) = child {
let _ = child.kill().await;
}
}
/// Waits for a child to print `READY <token>`, for at most READY_TIMEOUT.
/// Returns why it did not become ready otherwise.
async fn wait_for_child_ready(
stdout: tokio::process::ChildStdout,
token: &str,
abort_rx: &mut StopReceiver,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
) -> Result<BufReader<tokio::process::ChildStdout>, String> {
tokio::select! {
r = tokio::time::timeout(READY_TIMEOUT, wait_for_ready(stdout, token)) => match r {
Ok(Ok(reader)) => Ok(reader),
Ok(Err(e)) => Err(format!("did not become ready: {e}")),
Err(_) => Err(format!("not ready after {}s", READY_TIMEOUT.as_secs())),
},
// Any stop request, or appd shutting down, ends a startup.
_ = abort_rx.recv() => Err("startup aborted".to_owned()),
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
}
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
}
async fn wait_for_ready(
stdout: tokio::process::ChildStdout,
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
token: &str,
) -> anyhow::Result<BufReader<tokio::process::ChildStdout>> {
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
let expected = format!("READY {token}");
let mut reader = BufReader::new(stdout);
loop {
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
match read_child_line(&mut reader).await? {
None => return Err(anyhow::anyhow!("stdout closed without READY signal")),
// Earlier output without a trailing newline can precede the
// readiness line; the secret token alone establishes it.
Some(line) if line.ends_with(&expected) => return Ok(reader),
Some(_) => {}
}
}
}
async fn drain_stdout(mut reader: BufReader<tokio::process::ChildStdout>, session_id: u64) {
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
while let Ok(Some(line)) = read_child_line(&mut reader).await {
tracing::debug!(session_id, stdout = %line, "child stdout");
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
}
}
fix(appd): report sessions running only after the UI host is ready weft-appd marked a session Running and broadcast APP_READY as soon as the runtime printed READY, then started weft-app-shell with its output discarded and never watched it. A session with no UI, a UI host that failed to start, or one that crashed was still reported running. Early failure paths skipped parts of the cleanup: the image mount was not unmounted, the compositor association was not released, and the abort sender leaked on spawn failure. Readiness was any stdout line equal to READY, which a Wasm guest (inherited stdout) or a page (window.alert writes to the app shell's stdout) could print. Now a session becomes Running only after the runtime and then the app shell report readiness, each within 30 s; WEFT_APP_SHELL_BIN is required and checked before anything starts. After that, the runtime exiting, the app shell exiting or a terminate request ends the session. Every exit path goes through one teardown that kills both children, releases the compositor association, unmounts the image, stops the file portal and broadcasts Stopped. Readiness uses a per-session random token passed to both children in WEFT_READY_TOKEN. weft-runtime and weft-app-shell print READY <token>, and appd accepts only a line ending with it; guests receive only the variables the runtime names and pages cannot read the host environment. Child output is read in bounded, lossily decoded lines, so invalid UTF-8 or long lines no longer stop the readers. Compositor notifications use try_send so a disconnected compositor cannot stall teardown, and all children are killed if the supervisor is dropped. The IPC relay is not yet owned by the session, and children are killed without a graceful period. Validation: - cargo test -p weft-appd (43 passed); new tests cover: not ready until the app shell is ready, runtime exit before that, no app shell configured (runtime never starts), app shell exit ending a running session with the runtime process gone, and READY lines without the token; they fail against the previous supervisor - end to end with weft-compositor (nested), weft-appd, weft-runtime (wasmtime-runtime) and weft-app-shell: the Counter demo reaches APP_READY and is on screen; a page that prints a READY line through alert and never loads is stopped after 30 s without APP_READY; terminate leaves no child processes - clippy -D warnings and fmt --check for weft-appd and weft-runtime
2026-10-09 01:41:09 +00:00
async fn drain_stderr(stderr: tokio::process::ChildStderr, session_id: u64, process: &str) {
let mut reader = BufReader::new(stderr);
while let Ok(Some(line)) = read_child_line(&mut reader).await {
tracing::info!(session_id, process, stderr = %line, "child stderr");
feat(appd): implement runtime supervisor with process spawning and READY signal runtime.rs — process lifecycle manager: - supervise(session_id, app_id, registry): spawns the weft-runtime child process identified by WEFT_RUNTIME_BIN env var. If unset, logs debug and returns immediately (no-op until runtime binary is available). - Child process invoked as: <WEFT_RUNTIME_BIN> <app_id> <session_id> with stdout/stderr piped, stdin closed. - wait_for_ready(): reads stdout line-by-line; returns Ok(()) on first line matching 'READY'; returns Err if stdout closes without it. - 30-second READY_TIMEOUT via tokio::time::timeout; on expiry, kills the child and transitions session to Stopped. - On success: sets session state to Running, broadcasts AppReady to all connected WebSocket clients via registry broadcast channel. - drain_stderr(): async task that forwards child stderr lines to tracing at WARN level for observability. - On process exit: sets session state to Stopped regardless of exit code. main.rs — wiring: - SessionRegistry now owns broadcast::Sender<Response>; Default creates the channel internally. Added set_state(), subscribe(), broadcast() methods. Removed standalone broadcast_tx from run(); WS handlers subscribe via registry.lock().await.subscribe(). - dispatch::LaunchApp spawns a tokio task calling runtime::supervise immediately after creating the session. supervise is a no-op when WEFT_RUNTIME_BIN is unset, so existing tests are unaffected. Cargo.toml: added tokio 'process' and 'time' features.
2026-03-11 08:17:20 +00:00
}
}
#[cfg(all(test, target_os = "linux"))]
mod tests {
use super::SYSTEMD_SCOPE_ARGS;
/// systemd-run checks its options and their combinations before it
/// connects to the service manager. Run against an empty runtime
/// directory and bus address, it must get as far as the connection
/// attempt; any option error ends earlier with a different message. This
/// does not validate the `-p` property values, which only the manager
/// checks, and never starts a real scope.
#[test]
fn systemd_run_accepts_scope_options() {
let empty = std::env::temp_dir().join(format!("weft-scope-test-{}", std::process::id()));
std::fs::create_dir_all(&empty).unwrap();
let output = std::process::Command::new("systemd-run")
.args(SYSTEMD_SCOPE_ARGS)
.args(["--", "true"])
.env_clear()
.env("PATH", std::env::var_os("PATH").unwrap_or_default())
.env("LC_ALL", "C")
.env("XDG_RUNTIME_DIR", &empty)
.env(
"DBUS_SESSION_BUS_ADDRESS",
format!("unix:path={}/bus", empty.display()),
)
.output();
std::fs::remove_dir_all(&empty).unwrap();
let output = output.expect("systemd-run is required to check the scope options");
let stderr = String::from_utf8_lossy(&output.stderr);
// systemd 255 says "Failed to connect to bus"; later releases name the
// transport, e.g. "Failed to connect to user scope bus via local transport".
assert!(
stderr.starts_with("Failed to connect to"),
"systemd-run did not reach the manager connection: {stderr}"
);
}
}