Fix/daemon fixups

cli/internal/daemon/connector/connector.go

@@ -128,6 +128,7 @@ const (
 	_maxAttempts       = 10
 	_shutdownTimeout   = 1 * time.Second
 	_socketPollTimeout = 1 * time.Second
+	_notReadyTimeout   = 3 * time.Millisecond
 )
 
 // killLiveServer tells a running server to shut down. This method is also responsible
@@ -254,8 +255,11 @@ func (c *Connector) connectInternal(ctx context.Context) (*Client, error) {
 			}
 			// Loops back around and tries again.
 		} else if errors.Is(err, errUnavailable) {
-			// close the client, see if we can kill the stale daemon
+			// The rust daemon will open the socket a few ms before it's ready to accept connections.
+			// If we get here, we know that the socket exists, but the server isn't ready yet.
+			// We'll wait a few ms and try again.
 			c.Logger.Debug("server not ready yet")
+			time.Sleep(_notReadyTimeout)
 		} else if err != nil {
 			// Some other error occurred, close the client and
 			// report the error to the user
@@ -371,7 +375,7 @@ func (c *Connector) waitForSocket() error {
 
 // startDaemon starts the daemon and returns the pid for the new process
 func (c *Connector) startDaemon() (int, error) {
-	args := []string{"daemon"}
+	args := []string{"--skip-infer", "daemon"}
 	if c.Opts.ServerTimeout != 0 {
 		args = append(args, fmt.Sprintf("--idle-time=%v", c.Opts.ServerTimeout.String()))
 	}

crates/globwatch/src/lib.rs

@@ -62,7 +62,7 @@ impl GlobWatcher {
         std::fs::create_dir_all(&flush_dir).ok();
         let flush_dir = flush_dir.canonicalize()?;
 
-        let mut watcher = notify::recommended_watcher(move |event: Result<Event, Error>| {
+        let watcher = notify::recommended_watcher(move |event: Result<Event, Error>| {
             let span = span!(tracing::Level::TRACE, "watcher");
             let _ = span.enter();
 
@@ -83,10 +83,25 @@ impl GlobWatcher {
             }
         })?;
 
-        watcher.watch(flush_dir.as_path(), notify::RecursiveMode::Recursive)?;
-
         let watcher = Arc::new(Mutex::new(watcher));
 
+        // registering to watch this directory takes a few ms,
+        // so we just fire and forget a thread to do it in the
+        // background, to cut our startup time in half.
+        let flush = watcher.clone();
+        let path = flush_dir.as_path().to_owned();
+        tokio::task::spawn_blocking(move || {
+            if let Err(e) = flush
+                .lock()
+                .expect("only fails if poisoned")
+                .watch(&path, notify::RecursiveMode::Recursive)
+            {
+                warn!("failed to watch flush dir: {}", e);
+            } else {
+                trace!("watching flush dir: {:?}", path);
+            }
+        });
+
         Ok((
             Self {
                 flush_dir,

crates/pidlock/src/lib.rs

@@ -12,7 +12,7 @@ use log::warn;
 #[derive(Debug, thiserror::Error, PartialEq)]
 pub enum PidlockError {
     /// A lock already exists
-    #[error("lock exists at {0}")]
+    #[error("lock exists at \"{0}\", please remove it")]
     LockExists(PathBuf),
     /// An operation was attempted in the wrong state, e.g. releasing before
     /// acquiring.

crates/turborepo-lib/Cargo.toml

@@ -85,6 +85,7 @@ node-semver = "2.1.0"
 owo-colors.workspace = true
 regex.workspace = true
 tracing-appender = "0.2.2"
+tracing-chrome = { version = "0.7.1", optional = true }
 tracing-subscriber = { version = "0.3.17", features = ["env-filter"] }
 tracing.workspace = true
 turbo-updater = { workspace = true }

crates/turborepo-lib/src/commands/daemon.rs

@@ -1,11 +1,13 @@
 use std::{path::PathBuf, time::Duration};
 
+use pidlock::PidlockError::AlreadyOwned;
+use tracing::{trace, warn};
 use turbopath::{AbsoluteSystemPathBuf, RelativeSystemPathBuf};
 
 use super::CommandBase;
 use crate::{
     cli::DaemonCommand,
-    daemon::{DaemonConnector, DaemonError},
+    daemon::{endpoint::SocketOpenError, CloseReason, DaemonConnector, DaemonError},
     tracing::TurboSubscriber,
 };
 
@@ -65,6 +67,7 @@ pub async fn daemon_client(command: &DaemonCommand, base: &CommandBase) -> Resul
     Ok(())
 }
 
+#[tracing::instrument(skip(base, logging), fields(repo_root = %base.repo_root))]
 pub async fn daemon_server(
     base: &CommandBase,
     idle_time: &String,
@@ -101,7 +104,22 @@ pub async fn daemon_server(
         .map(|d| Duration::from_nanos(d as u64))?;
 
     let server = crate::daemon::DaemonServer::new(base, timeout, log_file)?;
-    server.serve().await;
+    let reason = server.serve().await;
+
+    match reason {
+        CloseReason::SocketOpenError(SocketOpenError::LockError(AlreadyOwned)) => {
+            warn!("daemon already running");
+        }
+        CloseReason::SocketOpenError(e) => return Err(e.into()),
+        CloseReason::Interrupt
+        | CloseReason::ServerClosed
+        | CloseReason::WatcherClosed
+        | CloseReason::Timeout
+        | CloseReason::Shutdown => {
+            // these are all ok, just exit
+            trace!("shutting down daemon: {:?}", reason);
+        }
+    };
 
     Ok(())
 }

crates/turborepo-lib/src/daemon/client.rs

@@ -3,7 +3,10 @@ use tonic::{Code, Status};
 use tracing::info;
 
 use self::proto::turbod_client::TurbodClient;
-use super::connector::{DaemonConnector, DaemonConnectorError};
+use super::{
+    connector::{DaemonConnector, DaemonConnectorError},
+    endpoint::SocketOpenError,
+};
 use crate::get_version;
 
 pub mod proto {
@@ -123,6 +126,8 @@ pub enum DaemonError {
     /// The server was connected but is now unavailable.
     #[error("server is unavailable")]
     Unavailable,
+    #[error("error opening socket: {0}")]
+    SocketOpen(#[from] SocketOpenError),
     /// The server is running a different version of turborepo.
     #[error("version mismatch")]
     VersionMismatch,

crates/turborepo-lib/src/daemon/connector.rs

@@ -148,6 +148,7 @@ impl DaemonConnector {
         // this creates a new process group for the given command
         // in a cross platform way, directing all output to /dev/null
         let mut group = tokio::process::Command::new(binary_path)
+            .arg("--skip-infer")
             .arg("daemon")
             .stderr(Stdio::null())
             .stdout(Stdio::null())

crates/turborepo-lib/src/daemon/endpoint.rs

@@ -5,14 +5,17 @@ use std::{io::ErrorKind, sync::atomic::Ordering, time::Duration};
 use futures::Stream;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tonic::transport::server::Connected;
-use tracing::debug;
+use tracing::{debug, trace};
 use turbopath::{AbsoluteSystemPathBuf, RelativeSystemPathBuf};
 
 #[derive(thiserror::Error, Debug)]
 pub enum SocketOpenError {
+    /// Returned when there is an IO error opening the socket,
+    /// such as the path being too long, or the path being
+    /// invalid.
     #[error("IO error: {0}")]
     Io(#[from] std::io::Error),
-    #[error("pidlock error")]
+    #[error("pidlock error: {0}")]
     LockError(#[from] pidlock::PidlockError),
 }
 
@@ -25,6 +28,7 @@ const WINDOWS_POLL_DURATION: Duration = Duration::from_millis(1);
 ///
 /// note: the running param is used by the windows
 ///       code path to shut down the non-blocking polling
+#[tracing::instrument]
 pub async fn listen_socket(
     path: AbsoluteSystemPathBuf,
     #[allow(unused)] running: Arc<AtomicBool>,
@@ -39,11 +43,12 @@ pub async fn listen_socket(
     let sock_path = path.join_relative(RelativeSystemPathBuf::new("turbod.sock").unwrap());
     let mut lock = pidlock::Pidlock::new(pid_path.as_path().to_owned());
 
+    trace!("acquiring pidlock");
     // this will fail if the pid is already owned
     lock.acquire()?;
     std::fs::remove_file(&sock_path).ok();
 
-    debug!("pidlock acquired as {}", pid_path);
+    debug!("pidlock acquired at {}", pid_path);
     debug!("listening on socket at {}", sock_path);
 
     #[cfg(unix)]

crates/turborepo-lib/src/daemon/mod.rs

@@ -7,7 +7,7 @@ mod server;
 
 pub use client::{DaemonClient, DaemonError};
 pub use connector::DaemonConnector;
-pub use server::DaemonServer;
+pub use server::{CloseReason, DaemonServer};
 
 pub(crate) mod proto {
     tonic::include_proto!("turbodprotocol");

crates/turborepo-lib/src/daemon/server.rs

@@ -71,6 +71,7 @@ pub enum CloseReason {
 }
 
 impl DaemonServer<notify::RecommendedWatcher> {
+    #[tracing::instrument(skip(base), fields(repo_root = %base.repo_root))]
     pub fn new(
         base: &CommandBase,
         timeout: Duration,
@@ -112,6 +113,7 @@ impl<T: Watcher> Drop for DaemonServer<T> {
 
 impl<T: Watcher + Send + 'static> DaemonServer<T> {
     /// Serve the daemon server, while also watching for filesystem changes.
+    #[tracing::instrument(skip(self))]
     pub async fn serve(mut self) -> CloseReason {
         let stop = StopSource::new();
         let watcher = self.watcher.clone();

crates/turborepo-lib/src/globwatcher/mod.rs

@@ -46,6 +46,7 @@ pub struct GlobSet {
 }
 
 impl HashGlobWatcher<RecommendedWatcher> {
+    #[tracing::instrument]
     pub fn new(
         relative_to: AbsoluteSystemPathBuf,
         flush_folder: PathBuf,
@@ -64,6 +65,7 @@ impl HashGlobWatcher<RecommendedWatcher> {
 impl<T: Watcher> HashGlobWatcher<T> {
     /// Watches a given path, using the flush_folder as temporary storage to
     /// make sure that file events are handled in the appropriate order.
+    #[tracing::instrument(skip(self, token))]
     pub async fn watch(&self, token: StopToken) {
         let start_globs = {
             let lock = self.hash_globs.lock().expect("only fails if poisoned");

crates/turborepo-lib/src/tracing.rs

@@ -5,7 +5,7 @@ use owo_colors::{
     colors::{Black, Default, Red, Yellow},
     Color, OwoColorize,
 };
-use tracing::{field::Visit, metadata::LevelFilter, Event, Level, Subscriber};
+use tracing::{field::Visit, metadata::LevelFilter, trace, Event, Level, Subscriber};
 use tracing_appender::{
     non_blocking::{NonBlocking, WorkerGuard},
     rolling::RollingFileAppender,
@@ -46,6 +46,9 @@ pub struct TurboSubscriber {
     /// The non-blocking file logger only continues to log while this guard is
     /// held. We keep it here so that it doesn't get dropped.
     guard: Mutex<Option<WorkerGuard>>,
+
+    #[cfg(feature = "tracing-chrome")]
+    chrome_guard: tracing_chrome::FlushGuard,
 }
 
 impl TurboSubscriber {
@@ -90,19 +93,33 @@ impl TurboSubscriber {
         // we set this layer to None to start with, effectively disabling it
         let (logrotate, update) = reload::Layer::new(Option::<DaemonLog>::None);
 
-        Registry::default().with(stdout).with(logrotate).init();
+        let registry = Registry::default().with(stdout).with(logrotate);
+
+        #[cfg(feature = "tracing-chrome")]
+        let (registry, chrome_guard) = {
+            let (chrome_layer, guard) = tracing_chrome::ChromeLayerBuilder::new()
+                .file("./tracing.json")
+                .build();
+            (registry.with(chrome_layer), guard)
+        };
+
+        registry.init();
 
         Self {
             update,
             guard: Mutex::new(None),
+            #[cfg(feature = "tracing-chrome")]
+            chrome_guard,
         }
     }
 
     /// Enables daemon logging with the specified rotation settings.
     ///
     /// Daemon logging uses the standard tracing formatter.
+    #[tracing::instrument(skip(self))]
     pub fn set_daemon_logger(&self, appender: RollingFileAppender) -> Result<(), Error> {
         let (file_writer, guard) = tracing_appender::non_blocking(appender);
+        trace!("created non-blocking file writer");
 
         let layer = tracing_subscriber::fmt::layer().with_writer(file_writer);
 

crates/turborepo/Cargo.toml

@@ -13,6 +13,7 @@ default = ["rustls-tls"]
 native-tls = ["turborepo-lib/native-tls"]
 rustls-tls = ["turborepo-lib/rustls-tls"]
 http = ["turborepo-lib/http"]
+tracing-chrome = ["turborepo-lib/tracing-chrome"]
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 [build-dependencies]

turborepo-tests/helpers/setup.sh

@@ -5,3 +5,4 @@ ROOT_DIR="${THIS_DIR}/../.."
 
 TURBO=${ROOT_DIR}/target/debug/turbo
 VERSION=${ROOT_DIR}/version.txt
+TMPDIR=$(mktemp -d)

turborepo-tests/integration/tests/daemon.t

@@ -0,0 +1,13 @@
+Setup
+  $ . ${TESTDIR}/../../helpers/setup.sh
+
+The daemon exits when there is a stale pid file
+  $ ${TURBO} daemon & sleep 1 && kill $! && ${TURBO} daemon
+  WARN stale pid file at ".+" (re)
+  ERROR error opening socket: pidlock error: lock exists at ".+", please remove it (re)
+  /bin/bash: line 4: .+ (re)
+  [1]
+
+A message is printed when the daemon is running already
+  $ rm -r ${TMPDIR}/turbod; ${TURBO} daemon & (export PID=$!; sleep 1 && ${TURBO} daemon && kill $PID && kill $PID && kill $PID)
+  WARN daemon already running