git.delta.rocks / fleet / refs/commits / 78196e185cf1

difftreelog

feat attic

ukmvyzywYaroslav Bolyukin2026-06-19parent: #2438e2b.patch.diff

22 files changed

modifiedCargo.lockdiffbeforeafterboth
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -5084,6 +5084,7 @@
  "remowt-ui-prompt",
  "tokio",
  "tracing",
+ "tracing-journald",
  "tracing-subscriber",
 ]
 
@@ -6576,6 +6577,17 @@
 ]
 
 [[package]]
+name = "tracing-journald"
+version = "0.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2d3a81ed245bfb62592b1e2bc153e77656d94ee6a0497683a65a12ccaf2438d0"
+dependencies = [
+ "libc",
+ "tracing-core",
+ "tracing-subscriber",
+]
+
+[[package]]
 name = "tracing-log"
 version = "0.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
modifiedCargo.tomldiffbeforeafterboth
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -75,6 +75,7 @@
 tokio = { version = "1.45.1", features = ["fs", "macros", "rt", "rt-multi-thread", "sync", "time"] }
 tracing = "0.1"
 tracing-indicatif = "0.3.13"
+tracing-journald = "0.3.2"
 tracing-opentelemetry = "0.33.0"
 uuid = { version = "1", features = ["v4"] }
 # For fixed coloring needs to be updated to https://github.com/tokio-rs/tracing/pull/3484
modifiedcmds/fleet/src/cmds/build_systems.rsdiffbeforeafterboth
--- a/cmds/fleet/src/cmds/build_systems.rs
+++ b/cmds/fleet/src/cmds/build_systems.rs
@@ -14,6 +14,8 @@
 use tokio::task::spawn_blocking;
 use tracing::{Instrument, error, field, info, info_span, warn};
 
+const TARGET: &str = "build-systems";
+
 #[derive(Parser)]
 pub struct Deploy {
 	/// Disable automatic rollback
@@ -32,7 +34,7 @@
 }
 
 async fn build_task(config: Config, hostname: String, build_attr: &str) -> Result<Utf8PathBuf> {
-	info!("building");
+	info!(target: TARGET, "building");
 	let host = config.host(&hostname)?;
 	// let action = Action::from(self.subcommand.clone());
 	let nixos = host.nixos_config()?;
@@ -43,7 +45,7 @@
 
 	// We already have system profiles for backups.
 	if !host.local {
-		info!("adding gc root");
+		info!(target: TARGET, "adding gc root");
 		let local = config.local_host();
 		let plugin_id = local.ensure_nix_plugin().await?;
 		let nix = local
@@ -70,7 +72,7 @@
 		let build_attr = self.build_attr.clone();
 		for host in hosts {
 			let config = config.clone();
-			let span = info_span!("build", host = field::display(&host.name));
+			let span = info_span!(target: TARGET, "build", host = field::display(&host.name));
 			let hostname = host.name;
 			let build_attr = build_attr.clone();
 			tasks.push(
@@ -78,7 +80,7 @@
 					let built = match build_task(config, hostname.clone(), &build_attr).await {
 						Ok(path) => path,
 						Err(e) => {
-							error!("failed to deploy host: {:?}", e);
+							error!(target: TARGET, "failed to deploy host: {:?}", e);
 							return;
 						}
 					};
@@ -86,9 +88,9 @@
 					let mut out = current_dir().expect("cwd exists");
 					out.push(format!("built-{hostname}"));
 
-					info!("linking iso image to {:?}", out);
+					info!(target: TARGET, "linking iso image to {:?}", out);
 					if let Err(e) = symlink(built, out) {
-						error!("failed to symlink: {e}")
+						error!(target: TARGET, "failed to symlink: {e}")
 					}
 				})
 				.instrument(span),
@@ -105,7 +107,7 @@
 		let tasks = FuturesUnordered::new();
 		for host in hosts.into_iter() {
 			let config = config.clone();
-			let span = info_span!("deploy", host = field::display(&host.name));
+			let span = info_span!(target: TARGET, "deploy", host = field::display(&host.name));
 			let hostname = host.name.clone();
 			let opts = opts.clone();
 			if let Some(deploy_kind) = opts.action_attr::<DeployKind>(&host, "deploy_kind")? {
@@ -125,7 +127,7 @@
 					{
 						Ok(path) => path,
 						Err(e) => {
-							error!("failed to build host system closure: {:?}", e);
+							error!(target: TARGET, "failed to build host system closure: {:?}", e);
 							return;
 						}
 					};
@@ -133,7 +135,7 @@
 					let deploy_kind = match host.deploy_kind().await {
 						Ok(v) => v,
 						Err(e) => {
-							error!("failed to query target deploy kind: {e}");
+							error!(target: TARGET, "failed to query target deploy kind: {e}");
 							return;
 						}
 					};
@@ -141,7 +143,7 @@
 					// TODO: Make disable_rollback a host attribute instead
 					let mut disable_rollback = self.disable_rollback;
 					if !disable_rollback && deploy_kind != DeployKind::Fleet {
-						warn!("disabling rollback, as not supported by non-fleet deployment kinds");
+						warn!(target: TARGET, "disabling rollback, as not supported by non-fleet deployment kinds");
 						disable_rollback = true;
 					}
 
@@ -149,7 +151,7 @@
 						match upload_task(&host, GenerationStorage::Deployer, built).await {
 							Ok(v) => v,
 							Err(e) => {
-								error!("upload failed: {e}");
+								error!(target: TARGET, "upload failed: {e}");
 								return;
 							}
 						};
@@ -169,7 +171,7 @@
 					)
 					.await
 					{
-						error!("activation failed: {e}");
+						error!(target: TARGET, "activation failed: {e}");
 					}
 				})
 				.instrument(span),
addedcmds/fleet/src/log_tree.rsdiffbeforeafterboth
--- /dev/null
+++ b/cmds/fleet/src/log_tree.rs
@@ -0,0 +1,243 @@
+//! Streaming, indented tree formatter.
+//!
+//! Unlike `tracing-tree` it keeps no per-thread depth state, so interleaved
+//! parallel output stays correct; unlike `tracing-forest` it doesn't buffer
+//! whole trees, so logs stream live. Visual style mirrors `@meteor-it/logger`'s
+//! node receiver: a fixed-width, right-aligned, level-colored target column,
+//! then a magenta `>`/`<` gutter for span enter/exit (with duration on exit),
+//! and 2-space indentation per depth for events.
+
+use std::{
+	fmt::Write as _,
+	io::{IsTerminal as _, Write as _},
+	sync::{
+		Mutex,
+		atomic::{AtomicBool, Ordering},
+	},
+	time::{Duration, Instant},
+};
+
+use tracing::{Event, Level, Subscriber, field::Field, field::Visit, span::Attributes, span::Id};
+use tracing_subscriber::{
+	fmt::MakeWriter,
+	layer::{Context, Layer},
+	registry::{LookupSpan, Scope},
+};
+
+/// Width of the right-aligned target column.
+const NAME_LIMIT: usize = 18;
+
+#[derive(Default)]
+struct Fields {
+	message: String,
+	rest: String,
+}
+
+impl Visit for Fields {
+	fn record_debug(&mut self, field: &Field, value: &dyn std::fmt::Debug) {
+		if field.name() == "message" {
+			let _ = write!(self.message, "{value:?}");
+		} else {
+			let _ = write!(self.rest, " {}={:?}", field.name(), value);
+		}
+	}
+}
+
+impl Fields {
+	fn render(&self) -> String {
+		format!("{}{}", self.message, self.rest).trim().to_owned()
+	}
+}
+
+/// Per-span state stashed in the registry's extensions.
+struct SpanData {
+	fields: String,
+	started: Instant,
+	// Whether the span's enter line was written, so we only emit an exit line
+	// for spans that actually appeared.
+	opened: AtomicBool,
+}
+
+/// SGR code → escape for the level-colored target column.
+fn level_code(level: &Level) -> &'static str {
+	match *level {
+		Level::ERROR => "31m", // red
+		Level::WARN => "33m",  // yellow
+		Level::INFO => "34m",  // blue
+		Level::DEBUG => "39m", // default fg
+		Level::TRACE => "90m", // gray
+	}
+}
+
+/// 2-space indent per depth, then a gutter symbol.
+fn gutter(depth: usize, sym: char) -> String {
+	let mut s = "  ".repeat(depth);
+	s.push(sym);
+	s
+}
+
+fn fmt_dur(d: Duration) -> String {
+	if d.as_secs_f64() >= 1.0 {
+		format!("{:.2}s", d.as_secs_f64())
+	} else if d.as_millis() >= 1 {
+		format!("{}ms", d.as_millis())
+	} else {
+		format!("{}µs", d.as_micros())
+	}
+}
+
+pub struct TreeLayer<W> {
+	writer: W,
+	color: bool,
+	// Serializes the whole compute-and-write so enter/exit decisions, output
+	// ordering, and the writes stay consistent under parallel emission.
+	last: Mutex<Vec<Id>>,
+}
+
+impl<W> TreeLayer<W> {
+	pub fn new(writer: W) -> Self {
+		Self {
+			writer,
+			color: std::io::stderr().is_terminal() && std::env::var_os("NO_COLOR").is_none(),
+			last: Mutex::new(vec![]),
+		}
+	}
+
+	/// Raw SGR escape, or empty when color is disabled.
+	fn sgr(&self, code: &str) -> String {
+		if self.color {
+			format!("\x1b[{code}")
+		} else {
+			String::new()
+		}
+	}
+
+	fn name_col(&self, mut target: &str, code: &str) -> String {
+		let w = NAME_LIMIT;
+		if target.len() > w {
+			target = &target[..w];
+		}
+		if self.color {
+			format!("\x1b[{code}\x1b[1m{target:>w$}\x1b[0m")
+		} else {
+			format!("{target:>w$}")
+		}
+	}
+
+	fn print_ancestors<S>(&self, scope: Option<Scope<'_, S>>, out: &mut String) -> usize
+	where
+		S: Subscriber + for<'a> LookupSpan<'a>,
+	{
+		let scope: Vec<_> = scope.map(|s| s.from_root().collect()).unwrap_or_default();
+
+		let mut _last = self.last.lock().expect("not poisoned");
+
+		let skip = scope
+			.iter()
+			.enumerate()
+			.take_while(|(i, s)| _last.get(*i) == Some(&s.id()))
+			.count();
+
+		for (depth, span) in scope.iter().enumerate().skip(skip) {
+			let ext = span.extensions();
+			let Some(data) = ext.get::<SpanData>() else {
+				continue;
+			};
+			let newly_open = !data.opened.swap(true, Ordering::Relaxed);
+			let _ = writeln!(
+				out,
+				"{} {}{}{} {}{}{}",
+				self.name_col(span.metadata().target(), "44m"),
+				self.sgr("35m"),
+				gutter(depth, if newly_open { '>' } else { '.' }),
+				self.sgr("1m"),
+				span.name(),
+				data.fields,
+				self.sgr("0m"),
+			);
+		}
+
+		*_last = scope.iter().map(|v| v.id()).collect();
+
+		scope.len()
+	}
+}
+
+impl<S, W> Layer<S> for TreeLayer<W>
+where
+	S: Subscriber + for<'a> LookupSpan<'a>,
+	W: for<'a> MakeWriter<'a> + 'static,
+{
+	fn on_new_span(&self, attrs: &Attributes<'_>, id: &Id, ctx: Context<'_, S>) {
+		let span = ctx.span(id).expect("new span exists");
+		let mut fields = Fields::default();
+		attrs.record(&mut fields);
+		span.extensions_mut().insert(SpanData {
+			fields: fields.rest,
+			started: Instant::now(),
+			opened: AtomicBool::new(false),
+		});
+	}
+
+	fn on_event(&self, event: &Event<'_>, ctx: Context<'_, S>) {
+		let mut out = String::new();
+
+		let scope = ctx.event_scope(event);
+		let pad = self.print_ancestors(scope, &mut out);
+
+		let mut fields = Fields::default();
+		event.record(&mut fields);
+		let message = fields.render();
+		let meta = event.metadata();
+		for (i, message) in message.split('\n').enumerate() {
+			// Event line: black bg + colored target column, then indent + message.
+			let _ = writeln!(
+				out,
+				"{}{}{}{}{}",
+				self.sgr("40m"),
+				self.name_col(
+					if i == 0 { meta.target() } else { "|" },
+					level_code(meta.level())
+				),
+				self.sgr("0m"),
+				gutter(pad, ' '),
+				message,
+			);
+		}
+
+		let _ = self.writer.make_writer().write_all(out.as_bytes());
+	}
+
+	fn on_close(&self, id: Id, ctx: Context<'_, S>) {
+		let Some(span) = ctx.span(&id) else { return };
+		let (opened, dur) = span
+			.extensions()
+			.get::<SpanData>()
+			.map_or((false, Duration::ZERO), |d| {
+				(d.opened.load(Ordering::Relaxed), d.started.elapsed())
+			});
+		if !opened {
+			return;
+		}
+
+		let mut out = String::new();
+		let _pad = self.print_ancestors(ctx.span_scope(&id), &mut out);
+
+		let depth = span.scope().skip(1).count(); // ancestors == depth
+		let _ = writeln!(
+			out,
+			"{} {}{}{} {}{} (Done in {}){}",
+			self.name_col(span.metadata().target(), "44m"),
+			self.sgr("35m"),
+			gutter(depth, '<'),
+			self.sgr("1m"),
+			span.name(),
+			self.sgr("22m"),
+			fmt_dur(dur),
+			self.sgr("0m"),
+		);
+
+		let _guard = self.last.lock().expect("not poisoned");
+		let _ = self.writer.make_writer().write_all(out.as_bytes());
+	}
+}
modifiedcmds/fleet/src/main.rsdiffbeforeafterboth
before · cmds/fleet/src/main.rs
1#![recursion_limit = "512"]23pub(crate) mod cmds;45use std::{process::ExitCode, sync::Arc};67use anyhow::{Context as _, Result, bail};8use camino::Utf8PathBuf;9use clap::{CommandFactory, Parser};10use cmds::{11	build_systems::{BuildSystems, Deploy},12	complete::Complete,13	info::Info,14	rollback::RollbackSingle,15	secrets::Secret,16	tf::Tf,17};18use fleet_base::{host::Config, opts::FleetOpts};19use futures::{TryStreamExt, stream::FuturesUnordered};20#[cfg(feature = "indicatif")]21use human_repr::HumanCount;22#[cfg(feature = "indicatif")]23use indicatif::{ProgressState, ProgressStyle};24use nix_eval::{25	eval_store, gc_register_my_thread, gc_unregister_my_thread, init_libraries, init_tokio_for_nix,26};27use opentelemetry::trace::TracerProvider;28use opentelemetry_appender_tracing::layer::OpenTelemetryTracingBridge;29use opentelemetry_exporter_env::{30	OtlpBaseSettings, OtlpLogsSettings, OtlpTracesSettings, ResolvedOtlpSettings,31};32use opentelemetry_sdk::{logs::SdkLoggerProvider, trace::SdkTracerProvider};33use tokio::task::spawn_blocking;34use tracing::{Instrument, error, info, info_span};35#[cfg(feature = "indicatif")]36use tracing_indicatif::IndicatifLayer;37use tracing_subscriber::{EnvFilter, prelude::*};3839#[derive(Parser)]40struct Prefetch {}41impl Prefetch {42	async fn run(&self, config: &Config) -> Result<()> {43		let mut prefetch_dir = config.directory.to_path_buf();44		prefetch_dir.push("prefetch");45		if !prefetch_dir.is_dir() {46			info!("nothing to prefetch: no prefetch directory");47			return Ok(());48		}49		let tasks = FuturesUnordered::new();50		for entry in std::fs::read_dir(&prefetch_dir)? {51			let entry = entry?;52			if !entry.metadata()?.is_file() {53				bail!("only files should exist in prefetch directory");54			}55			let name = entry.file_name().to_string_lossy().into_owned();56			let path =57				Utf8PathBuf::try_from(entry.path()).context("prefetch path should be utf8")?;58			let span = info_span!("prefetching", name = %name);59			tasks.push(async move {60				let store = eval_store();61				let added = spawn_blocking(move || store.add_file(&name, &path))62					.instrument(span.clone())63					.await??;64				let _g = span.enter();65				info!("{} -> {}", added.hash, added.store_path);66				anyhow::Ok(())67			});68		}69		tasks.try_collect::<Vec<()>>().await?;70		Ok(())71	}72}7374#[derive(Parser)]75enum Opts {76	/// Build system closures77	BuildSystems(BuildSystems),78	/// Upload and switch system closures79	Deploy(Deploy),80	/// Rollback remote machine by redeploying old generation as the new one81	RollbackSingle(RollbackSingle),82	/// Secret management83	#[clap(subcommand)]84	Secret(Secret),85	/// Upload prefetch directory to the nix store86	Prefetch(Prefetch),87	/// Config parsing88	Info(Info),89	/// Command completions90	#[clap(hide(true))]91	Complete(Complete),92	/// Compile and evaluate terranix configuration93	Tf(Tf),94}9596#[derive(Parser)]97#[clap(version, author)]98struct RootOpts {99	#[clap(flatten)]100	fleet_opts: FleetOpts,101	#[clap(subcommand)]102	command: Opts,103	#[clap(long, next_help_heading = "Telemetry", env = "OTEL_FLEET")]104	otel: bool,105	#[clap(flatten)]106	otlp_base: OtlpBaseSettings,107	#[clap(flatten)]108	otel_logs: OtlpLogsSettings,109	#[clap(flatten)]110	otel_traces: OtlpTracesSettings,111}112113async fn run_command(config: &Config, opts: FleetOpts, command: Opts) -> Result<()> {114	match command {115		Opts::BuildSystems(c) => c.run(config, &opts).await?,116		Opts::Deploy(d) => d.run(config, &opts).await?,117		Opts::RollbackSingle(r) => r.run(config, &opts).await?,118		Opts::Secret(s) => s.run(config, &opts).await?,119		Opts::Info(i) => i.run(config).await?,120		Opts::Prefetch(p) => p.run(config).await?,121		Opts::Tf(t) => t.run(config).await?,122		// TODO: actually parse commands before starting the async runtime123		Opts::Complete(c) => spawn_blocking(move || c.run(RootOpts::command())).await?,124	};125	Ok(())126}127128fn setup_logging(opts: &RootOpts) -> Result<()> {129	#[cfg(feature = "indicatif")]130	let indicatif_layer = {131		use std::time::Duration;132133		IndicatifLayer::new().with_max_progress_bars(10, Some(ProgressStyle::default_spinner()))134			.with_progress_style(135			ProgressStyle::with_template(136				"{color_start}{span_child_prefix} {span_name}{{{span_fields}}}{color_end} {wide_msg} {color_start}{download_progress} {elapsed}{color_end}",137			)138				.unwrap()139				.with_key("download_progress", |state: &ProgressState, writer: &mut dyn std::fmt::Write| {140					let Some(len) = state.len() else {141						return;142					};143					let pos = state.pos();144					if pos > len {145						let _ = write!(writer, "{}", pos.human_count_bare());146					} else {147						let _ = write!(writer, "{} / {}", pos.human_count_bare(), len.human_count_bare());148					}149				})150				.with_key(151					"color_start",152					|state: &ProgressState, writer: &mut dyn std::fmt::Write| {153						let elapsed = state.elapsed();154155						if elapsed > Duration::from_secs(60) {156							// Red157							let _ = write!(writer, "\x1b[{}m", 1 + 30);158						} else if elapsed > Duration::from_secs(30) {159							// Yellow160							let _ = write!(writer, "\x1b[{}m", 3 + 30);161						}162					},163				)164				.with_key(165					"color_end",166					|state: &ProgressState, writer: &mut dyn std::fmt::Write| {167						if state.elapsed() > Duration::from_secs(30) {168							let _ = write!(writer, "\x1b[0m");169						}170					},171				),172		)173	};174175	let filter = EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new("info"));176177	let reg = tracing_subscriber::registry().with({178		let sub = tracing_subscriber::fmt::layer().without_time();179		#[cfg(feature = "indicatif")]180		let sub = sub.with_writer(indicatif_layer.get_stderr_writer());181		sub.with_filter(filter) // .without,182	});183184	#[cfg(feature = "indicatif")]185	let reg = reg.with(indicatif_layer);186187	if opts.otel {188		let traces = ResolvedOtlpSettings::traces(&opts.otlp_base, &opts.otel_traces)?;189		let span_exporter = traces.span_exporter()?;190		let logs = ResolvedOtlpSettings::logs(&opts.otlp_base, &opts.otel_logs)?;191		let log_exporter = logs.log_exporter()?;192193		let span_provider = SdkTracerProvider::builder()194			.with_batch_exporter(span_exporter)195			.build();196		let log_provider = SdkLoggerProvider::builder()197			.with_batch_exporter(log_exporter)198			.build();199200		let logger = OpenTelemetryTracingBridge::new(&log_provider);201		let tracer = span_provider.tracer("fleet");202203		let reg = reg204			.with(tracing_opentelemetry::layer().with_tracer(tracer))205			.with(logger);206207		reg.init();208	} else {209		reg.init();210	};211212	Ok(())213}214215fn main() -> ExitCode {216	let opts = RootOpts::parse();217	if let Opts::Complete(c) = &opts.command {218		c.run(RootOpts::command());219		return ExitCode::SUCCESS;220	}221222	if let Err(e) = setup_logging(&opts) {223		eprintln!("{e:#}");224		return ExitCode::FAILURE;225	}226227	init_libraries();228229	let runtime = tokio::runtime::Builder::new_multi_thread()230		.enable_all()231		.on_thread_start(|| {232			gc_register_my_thread();233		})234		.on_thread_stop(|| {235			gc_unregister_my_thread();236		})237		.build()238		.expect("failed to build runtime");239	let runtime = Arc::new(runtime);240241	init_tokio_for_nix(runtime.clone());242243	runtime.block_on(async {244		tokio::task::spawn(async move {245			if let Err(e) = main_real(opts).await {246				error!("{e:#}");247				ExitCode::FAILURE248			} else {249				ExitCode::SUCCESS250			}251		})252		.await253		.expect("primary task panicked")254	})255	// async_main(opts)256}257258async fn main_real(opts: RootOpts) -> Result<()> {259	let config = opts.fleet_opts.build(matches!(260		opts.command,261		Opts::Deploy(_) | Opts::BuildSystems(_)262	))?;263264	match run_command(&config, opts.fleet_opts, opts.command).await {265		Ok(()) => {266			config.save()?;267			Ok(())268		}269		Err(e) => {270			let _ = config.save();271			Err(e)272		}273	}274}275276#[cfg(test)]277mod tests {278	use super::*;279280	#[test]281	fn verify_command() {282		use clap::CommandFactory;283		RootOpts::command().debug_assert();284	}285}
after · cmds/fleet/src/main.rs
1#![recursion_limit = "512"]23pub(crate) mod cmds;4mod log_tree;56use std::{process::ExitCode, sync::Arc};78use anyhow::{Context as _, Result, bail};9use camino::Utf8PathBuf;10use clap::{CommandFactory, Parser};11use cmds::{12	build_systems::{BuildSystems, Deploy},13	complete::Complete,14	info::Info,15	rollback::RollbackSingle,16	secrets::Secret,17	tf::Tf,18};19use fleet_base::{host::Config, opts::FleetOpts};20use futures::{TryStreamExt, stream::FuturesUnordered};21#[cfg(feature = "indicatif")]22use human_repr::HumanCount;23#[cfg(feature = "indicatif")]24use indicatif::{ProgressState, ProgressStyle};25use log_tree::TreeLayer;26use nix_eval::{27	eval_store, gc_register_my_thread, gc_unregister_my_thread, init_libraries, init_tokio_for_nix,28};29use opentelemetry::trace::TracerProvider;30use opentelemetry_appender_tracing::layer::OpenTelemetryTracingBridge;31use opentelemetry_exporter_env::{32	OtlpBaseSettings, OtlpLogsSettings, OtlpTracesSettings, ResolvedOtlpSettings,33};34use opentelemetry_sdk::{logs::SdkLoggerProvider, trace::SdkTracerProvider};35use tokio::task::spawn_blocking;36use tracing::{Instrument, error, info, info_span};37#[cfg(feature = "indicatif")]38use tracing_indicatif::IndicatifLayer;39use tracing_subscriber::{EnvFilter, prelude::*};4041#[derive(Parser)]42struct Prefetch {}43impl Prefetch {44	async fn run(&self, config: &Config) -> Result<()> {45		let mut prefetch_dir = config.directory.to_path_buf();46		prefetch_dir.push("prefetch");47		if !prefetch_dir.is_dir() {48			info!("nothing to prefetch: no prefetch directory");49			return Ok(());50		}51		let tasks = FuturesUnordered::new();52		for entry in std::fs::read_dir(&prefetch_dir)? {53			let entry = entry?;54			if !entry.metadata()?.is_file() {55				bail!("only files should exist in prefetch directory");56			}57			let name = entry.file_name().to_string_lossy().into_owned();58			let path =59				Utf8PathBuf::try_from(entry.path()).context("prefetch path should be utf8")?;60			let span = info_span!("prefetching", name = %name);61			tasks.push(async move {62				let store = eval_store();63				let added = spawn_blocking(move || store.add_file(&name, &path))64					.instrument(span.clone())65					.await??;66				let _g = span.enter();67				info!("{} -> {}", added.hash, added.store_path);68				anyhow::Ok(())69			});70		}71		tasks.try_collect::<Vec<()>>().await?;72		Ok(())73	}74}7576#[derive(Parser)]77enum Opts {78	/// Build system closures79	BuildSystems(BuildSystems),80	/// Upload and switch system closures81	Deploy(Deploy),82	/// Rollback remote machine by redeploying old generation as the new one83	RollbackSingle(RollbackSingle),84	/// Secret management85	#[clap(subcommand)]86	Secret(Secret),87	/// Upload prefetch directory to the nix store88	Prefetch(Prefetch),89	/// Config parsing90	Info(Info),91	/// Command completions92	#[clap(hide(true))]93	Complete(Complete),94	/// Compile and evaluate terranix configuration95	Tf(Tf),96}9798#[derive(Parser)]99#[clap(version, author)]100struct RootOpts {101	#[clap(flatten)]102	fleet_opts: FleetOpts,103	#[clap(subcommand)]104	command: Opts,105	#[clap(long, next_help_heading = "Telemetry", env = "OTEL_FLEET")]106	otel: bool,107	#[clap(flatten)]108	otlp_base: OtlpBaseSettings,109	#[clap(flatten)]110	otel_logs: OtlpLogsSettings,111	#[clap(flatten)]112	otel_traces: OtlpTracesSettings,113}114115async fn run_command(config: &Config, opts: FleetOpts, command: Opts) -> Result<()> {116	match command {117		Opts::BuildSystems(c) => c.run(config, &opts).await?,118		Opts::Deploy(d) => d.run(config, &opts).await?,119		Opts::RollbackSingle(r) => r.run(config, &opts).await?,120		Opts::Secret(s) => s.run(config, &opts).await?,121		Opts::Info(i) => i.run(config).await?,122		Opts::Prefetch(p) => p.run(config).await?,123		Opts::Tf(t) => t.run(config).await?,124		// TODO: actually parse commands before starting the async runtime125		Opts::Complete(c) => spawn_blocking(move || c.run(RootOpts::command())).await?,126	};127	Ok(())128}129130fn setup_logging(opts: &RootOpts) -> Result<()> {131	#[cfg(feature = "indicatif")]132	let indicatif_layer = {133		use std::fmt;134		use std::time::Duration;135136		IndicatifLayer::new().with_max_progress_bars(10, Some(ProgressStyle::default_spinner()))137			.with_span_child_prefix_indent("  ")138			.with_span_child_prefix_symbol("")139			.with_progress_style(140			ProgressStyle::with_template(141				"{span_child_prefix:.magenta}{chevron:.magenta}{span_name:<18.blue.bold} {wide_msg} {span_fields:.dim} {color_start}{download_progress} {elapsed}{color_end}",142			)143				.unwrap()144				.with_key("chevron", |_: &ProgressState, writer: &mut dyn fmt::Write| {145					let _ = write!(writer, "> ");146				})147				.with_key("download_progress", |state: &ProgressState, writer: &mut dyn fmt::Write| {148					let Some(len) = state.len() else {149						return;150					};151					let pos = state.pos();152					if pos > len {153						let _ = write!(writer, "{}", pos.human_count_bare());154					} else {155						let _ = write!(writer, "{} / {}", pos.human_count_bare(), len.human_count_bare());156					}157				})158				.with_key(159					"color_start",160					|state: &ProgressState, writer: &mut dyn fmt::Write| {161						let elapsed = state.elapsed();162163						if elapsed > Duration::from_secs(60) {164							// Red165							let _ = write!(writer, "\x1b[{}m", 1 + 30);166						} else if elapsed > Duration::from_secs(30) {167							// Yellow168							let _ = write!(writer, "\x1b[{}m", 3 + 30);169						}170					},171				)172				.with_key(173					"color_end",174					|state: &ProgressState, writer: &mut dyn fmt::Write| {175						if state.elapsed() > Duration::from_secs(30) {176							let _ = write!(writer, "\x1b[0m");177						}178					},179				),180		)181	};182183	let filter = EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new("info"));184185	let tree = {186		#[cfg(feature = "indicatif")]187		let writer = indicatif_layer.get_stderr_writer();188		#[cfg(not(feature = "indicatif"))]189		let writer = || std::io::stderr();190		TreeLayer::new(writer)191	};192193	let reg = tracing_subscriber::registry().with(tree.with_filter(filter));194195	#[cfg(feature = "indicatif")]196	let reg = reg.with(indicatif_layer);197198	if opts.otel {199		let traces = ResolvedOtlpSettings::traces(&opts.otlp_base, &opts.otel_traces)?;200		let span_exporter = traces.span_exporter()?;201		let logs = ResolvedOtlpSettings::logs(&opts.otlp_base, &opts.otel_logs)?;202		let log_exporter = logs.log_exporter()?;203204		let span_provider = SdkTracerProvider::builder()205			.with_batch_exporter(span_exporter)206			.build();207		let log_provider = SdkLoggerProvider::builder()208			.with_batch_exporter(log_exporter)209			.build();210211		let logger = OpenTelemetryTracingBridge::new(&log_provider);212		let tracer = span_provider.tracer("fleet");213214		reg.with(tracing_opentelemetry::layer().with_tracer(tracer))215			.with(logger)216			.init();217	} else {218		reg.init();219	}220221	Ok(())222}223224fn main() -> ExitCode {225	let opts = RootOpts::parse();226	if let Opts::Complete(c) = &opts.command {227		c.run(RootOpts::command());228		return ExitCode::SUCCESS;229	}230231	if let Err(e) = setup_logging(&opts) {232		eprintln!("{e:#}");233		return ExitCode::FAILURE;234	}235236	init_libraries();237238	let runtime = tokio::runtime::Builder::new_multi_thread()239		.enable_all()240		.on_thread_start(|| {241			gc_register_my_thread();242		})243		.on_thread_stop(|| {244			gc_unregister_my_thread();245		})246		.build()247		.expect("failed to build runtime");248	let runtime = Arc::new(runtime);249250	init_tokio_for_nix(runtime.clone());251252	runtime.block_on(async {253		tokio::task::spawn(async move {254			if let Err(e) = main_real(opts).await {255				error!("{e:#}");256				ExitCode::FAILURE257			} else {258				ExitCode::SUCCESS259			}260		})261		.await262		.expect("primary task panicked")263	})264}265266async fn main_real(opts: RootOpts) -> Result<()> {267	let config = opts.fleet_opts.build(matches!(268		opts.command,269		Opts::Deploy(_) | Opts::BuildSystems(_)270	))?;271272	match run_command(&config, opts.fleet_opts, opts.command).await {273		Ok(()) => {274			config.save()?;275			Ok(())276		}277		Err(e) => {278			let _ = config.save();279			Err(e)280		}281	}282}283284#[cfg(test)]285mod tests {286	use super::*;287288	#[test]289	fn verify_command() {290		use clap::CommandFactory;291		RootOpts::command().debug_assert();292	}293}
modifiedcrates/fleet-base/src/deploy.rsdiffbeforeafterboth
--- a/crates/fleet-base/src/deploy.rs
+++ b/crates/fleet-base/src/deploy.rs
@@ -290,8 +290,10 @@
 	if !host.local {
 		info!("uploading system closure");
 		let mut tries = 0;
+		// TODO: Use gc prefix?..
 		loop {
-			match host.remote_derivation(&generation).await {
+			let name = format!("{}-system", host.gc_root_prefix());
+			match host.remote_derivation(&generation, Some(&name)).await {
 				Ok(remote) => {
 					assert!(remote == generation, "CA derivations aren't implemented");
 					return Ok(remote);
modifiedcrates/fleet-base/src/host.rsdiffbeforeafterboth
--- a/crates/fleet-base/src/host.rs
+++ b/crates/fleet-base/src/host.rs
@@ -13,16 +13,16 @@
 use camino::{Utf8Path, Utf8PathBuf};
 use chrono::{DateTime, Utc};
 use fleet_shared::SecretData;
-use nix_eval::{Store, Value, drv::DrvGraph, eval_store, nix_go, nix_go_json, util::assert_warn};
+use nix_eval::{Store, Value, eval_store, nix_go, nix_go_json, util::assert_warn};
 use remowt_client::{AgentBundle, Remowt};
 use remowt_endpoints::fs::FsClient;
 use remowt_fleet::NixClient;
-use remowt_link_shared::BifConfig;
+use remowt_link_shared::{Address, BifConfig};
 use remowt_ui_prompt::auto::AutoPrompter;
 use remowt_ui_prompt::bifrost::PromptEndpoints;
 use remowt_ui_prompt::{PrependSourcePrompter, Source};
 use tempfile::NamedTempFile;
-use tokio::{sync::OnceCell, task::spawn_blocking};
+use tokio::{process::Command, sync::OnceCell, task::spawn_blocking};
 use tracing::{info, instrument, warn};
 
 use crate::fleetdata::{
@@ -104,11 +104,14 @@
 	session_destination: OnceLock<String>,
 	legacy_ssh_store: OnceLock<bool>,
 
+	/// fleetConfiguration.hosts.host, None for local (as local is only used for local tool overrides)
 	pub host_config: Option<Value>,
 	pub nixos_config: OnceLock<Value>,
 	pub nixos_unchecked_config: OnceLock<Value>,
 	pub pkgs_override: Option<Value>,
 
+	binary_cache: OnceCell<Option<BinaryCache>>,
+
 	// TODO: Move command helpers away with connectivity refactor
 	pub local: bool,
 	pub remowt: OnceCell<Remowt>,
@@ -131,13 +134,35 @@
 	AgentBundle::from_dir(agents_dir()?)
 }
 
-enum RemoteDerivationMode {
-	/// Closure is (pre)signed, remote host is expected to trust our signatures.
-	CopySigned,
-	/// Closure is not signed, using escalated nix daemon to bypass signature verification.
-	CopyPrivileged,
-	/// Closure is uploaded to attic, remote host is expected to trust its signature.
-	Attic { name: String },
+#[derive(Clone)]
+pub enum BinaryCache {
+	Attic { name: String, pin: bool },
+}
+impl BinaryCache {
+	async fn upload(&self, local: &Remowt, path: &Utf8Path, pin_name: Option<&str>) -> Result<()> {
+		match self {
+			BinaryCache::Attic { name, pin } => {
+				let mut cmd = local.cmd("attic");
+				cmd.arg("push").arg(name).arg(path.as_str());
+				cmd.run()
+					.await
+					.context("failed to push path to attic (is attic-cli on PATH?)")?;
+
+				if *pin && let Some(pin_name) = pin_name {
+					let mut cmd = local.cmd("attic");
+					cmd.arg("pin")
+						.arg("create")
+						.arg(name)
+						.arg(pin_name)
+						.arg(path.as_str());
+					cmd.run()
+						.await
+						.context("failed to pin path in attic (is your attic-cli compatible?)")?;
+				}
+				Ok(())
+			}
+		}
+	}
 }
 
 impl ConfigHost {
@@ -156,6 +181,32 @@
 			.set(legacy)
 			.expect("legacy ssh store is already set")
 	}
+	pub fn binary_cache(&self) -> Result<Option<BinaryCache>> {
+		if let Some(v) = self.binary_cache.get() {
+			return Ok(v.clone());
+		}
+		let cache = if let Some(host_config) = &self.host_config {
+			let cache = nix_go!(host_config.cache);
+			let enable: bool = nix_go_json!(cache.enable);
+			if enable {
+				let kind: String = nix_go_json!(cache.kind);
+				let name: String = nix_go_json!(cache.name);
+				match kind.as_str() {
+					"attic" => {
+						let pin: bool = nix_go_json!(cache.pin);
+						Some(BinaryCache::Attic { name, pin })
+					}
+					v => bail!("unknown binary cache kind: {v}"),
+				}
+			} else {
+				None
+			}
+		} else {
+			None
+		};
+		let _ = self.binary_cache.set(cache.clone());
+		Ok(cache)
+	}
 	pub async fn deploy_kind(&self) -> Result<DeployKind> {
 		self.deploy_kind.get_or_try_init(|| async {
 			let remowt = self.remowt().await?;
@@ -238,8 +289,10 @@
 					let built = plugin
 						.build("out")
 						.context("failed to build the fleet nix plugin")?;
+					let pin = format!("{}-remowt-plugin-fleet", self.gc_root_prefix());
 					let copied = self
-						.remote_derivation(&built)
+						// TODO: Pin? But this pin will be host-specific, and heavy because of nix dependency...
+						.remote_derivation(&built, Some(&pin))
 						.await
 						.context("failed to copy the fleet nix plugin to the host store")?;
 					let bin = copied.join("bin/remowt-plugin-fleet");
@@ -248,6 +301,13 @@
 						.run0_load_plugin_path(NIX_PLUGIN_ID, bin.as_str())
 						.await
 						.context("failed to load the fleet nix plugin")?;
+					// Should not fail, as plugin loading waits for connection on remote, and we only wait for forward... On the other hand, we should monitor connection to the remote itself somewhere too.
+					self.remowt()
+						.await?
+						.rpc()
+						.wait_for_connection_to(Address::Plugin(NIX_PLUGIN_ID))
+						.await
+						.map_err(|_| anyhow!("failed to wait"))?;
 					anyhow::Ok(())
 				})
 				.await?;
@@ -339,17 +399,31 @@
 	}
 	/// Returns path for futureproofing, as path might change i.e on conversion to CA
 	#[instrument(skip(self))]
-	pub async fn remote_derivation(&self, path: &Utf8Path) -> Result<Utf8PathBuf> {
+	pub async fn remote_derivation(
+		&self,
+		path: &Utf8Path,
+		pin_name: Option<&str>,
+	) -> Result<Utf8PathBuf> {
 		let path = path.to_owned();
 		if self.local {
 			// Path is located locally, thus already trusted.
 			return Ok(path);
 		}
+		let cache = self.binary_cache()?;
+		let mut cache = if let Some(cache) = cache {
+			let local = self.config.local_host().remowt().await?;
+			let path = path.to_owned();
+			let pin_name = pin_name.map(str::to_owned);
+			Some(tokio::spawn(async move {
+				if let Err(e) = cache.upload(&local, &path, pin_name.as_deref()).await {
+					warn!("failed to upload closure {path} to cache: {e:?}");
+				}
+			}))
+		} else {
+			None
+		};
 		let sign: Pin<Box<dyn Future<Output = Result<()>> + Send>> = {
 			let path = path.clone();
-			let graph = DrvGraph::resolve(&eval_store(), &path)
-				.context("failed to resolve graph to be uploaded")?;
-			info!("signing {} paths", graph.nodes.len());
 
 			Box::pin(async move {
 				let local = self.config.local_host();
@@ -377,13 +451,62 @@
 				.expect("copy_to panicked")
 				.context("copying closure to remote store")?;
 		}
+
+		if let Some(cache) = cache.take() {
+			let _ = cache.await;
+		}
+
 		Ok(path)
 	}
+
+	/// Like [`Self::remote_derivation`], but instead of copying the closure over
+	/// the nix daemon tunnel, uploads it to an attic binary cache and has the
+	/// remote host substitute it from there. The remote is expected to already
+	/// trust the cache's signing key and have it configured as a substituter.
+	#[instrument(skip(self))]
+	pub async fn remote_derivation_attic(
+		&self,
+		path: &Utf8Path,
+		cache_name: &str,
+	) -> Result<Utf8PathBuf> {
+		let path = path.to_owned();
+		if self.local {
+			// Path is located locally, thus already trusted.
+			return Ok(path);
+		}
+
+		info!("uploading {path} closure to attic cache {cache_name}");
+		let status = Command::new("attic")
+			.arg("push")
+			.arg(cache_name)
+			.arg(path.as_str())
+			.status()
+			.await
+			.context("failed to spawn attic-cli, is it on PATH?")?;
+		ensure!(status.success(), "attic push failed: {status}");
+
+		info!("substituting {path} on the remote host");
+		let nix = self.nix_client().await?;
+		let substituted = nix
+			.substitute(vec![path.clone()])
+			.await
+			.map_err(|e| anyhow!("{e:?}"))?
+			.map_err(|e| anyhow!("{e}"))?;
+		ensure!(
+			substituted.contains(&path),
+			"remote host failed to substitute {path} from attic cache {cache_name}"
+		);
+		Ok(path)
+	}
 }
 
 struct HostSecretDefinition(Value);
 
 impl ConfigHost {
+	pub fn gc_root_prefix(&self) -> String {
+		format!("{}-{}", self.config.gc_root_prefix(), self.name)
+	}
+
 	// TOCTOU is possible here in case if config is changed, but this case is not handled anywhere anyway,
 	// assuming getting tags always returns the same value.
 	pub fn tags(&self) -> Result<Vec<String>> {
@@ -504,6 +627,7 @@
 						cell
 					},
 					pkgs_override: Some(self.default_pkgs.clone()),
+					binary_cache: OnceCell::new(),
 
 					local: true,
 					remowt: OnceCell::new(),
@@ -546,6 +670,7 @@
 			nixos_unchecked_config: OnceLock::new(),
 			groups: OnceLock::new(),
 			pkgs_override: None,
+			binary_cache: OnceCell::new(),
 
 			// TODO: Remove with connectivit refactor
 			local: self.localhost == name,
modifiedcrates/fleet-base/src/keys.rsdiffbeforeafterboth
--- a/crates/fleet-base/src/keys.rs
+++ b/crates/fleet-base/src/keys.rs
@@ -9,6 +9,9 @@
 use crate::{fleetdata::SecretOwner, host::Config};
 
 impl Config {
+	pub fn gc_root_prefix(&self) -> String {
+		self.data.gc_root_prefix.clone()
+	}
 	fn cached_host_key(&self, host: &str) -> Option<String> {
 		let hosts = self.data.hosts.read().expect("no poisoning");
 		let key = hosts.get(host).map(|h| &h.encryption_key);
modifiedcrates/fleet-base/src/primops.rsdiffbeforeafterboth
--- a/crates/fleet-base/src/primops.rs
+++ b/crates/fleet-base/src/primops.rs
@@ -119,7 +119,7 @@
 				.context("failed to build generator for target host")?;
 
 			let generator = host_on
-				.remote_derivation(&generator)
+				.remote_derivation(&generator, None)
 				.await
 				.context("failed to copy generator to target host")?;
 
modifiedcrates/nix-eval/src/lib.rsdiffbeforeafterboth
--- a/crates/nix-eval/src/lib.rs
+++ b/crates/nix-eval/src/lib.rs
@@ -39,7 +39,7 @@
 	make_attrs, make_bindings_builder, make_list, make_list_builder, realised_string,
 	realised_string_free, realised_string_get_buffer_size, realised_string_get_buffer_start,
 	realised_string_get_store_path, realised_string_get_store_path_count, register_primop,
-	set_err_msg, setting_set, state_free, store_copy_closure, store_free, store_open,
+	set_err_msg, setting_get, setting_set, state_free, store_copy_closure, store_free, store_open,
 	store_parse_path, store_path_free, store_path_name, string_realise, value, value_call,
 	value_decref, value_force, value_incref,
 };
@@ -390,6 +390,14 @@
 	with_default_context(|c, _| unsafe { setting_set(c, s.as_ptr(), v.as_ptr()) }).map(|_| ())
 }
 
+pub fn get_setting(s: &CStr) -> Result<String> {
+	let mut out = String::new();
+	with_default_context(|c, _| unsafe {
+		setting_get(c, s.as_ptr(), Some(copy_nix_str), (&raw mut out).cast())
+	})?;
+	Ok(out)
+}
+
 #[derive(Debug)]
 pub struct AddedFile {
 	pub store_path: Utf8PathBuf,
modifiedcrates/nix-eval/src/logging.rsdiffbeforeafterboth
--- a/crates/nix-eval/src/logging.rs
+++ b/crates/nix-eval/src/logging.rs
@@ -109,6 +109,12 @@
 			(ActivityType::CopyPaths, []) => {
 				debug_span!(target: "nix::copy-paths", "copying paths")
 			}
+			// Umbrella progress activity. Nix emits it at lvlError ("always show"),
+			// which would otherwise become a noisy ERROR `action` span — give it an
+			// explicit debug level like the other container activities.
+			(ActivityType::Builds, []) => {
+				debug_span!(target: "nix::builds", "building")
+			}
 			(ActivityType::Unknown, [])
 				if s.starts_with("copying \"") && s.ends_with("\" to the store") =>
 			{
@@ -365,6 +371,16 @@
 	BuildGraphGuard { paths }
 }
 
+/// Drop the lazily-created `building` span for a derivation. Used when a build
+/// fails: its dependents will never build, so their speculatively-created spans
+/// (made when a dependency started building) would otherwise linger until the
+/// whole build graph is torn down.
+pub fn remove_drv_span(drv_path: &Utf8Path) {
+	if let Some(entry) = DRV_GRAPH.lock().expect("not poisoned").get_mut(drv_path) {
+		entry.span = None;
+	}
+}
+
 fn ensure_drv_span(drv_path: &Utf8Path) -> Option<Span> {
 	let mut drv_graph = DRV_GRAPH.lock().expect("not poisoned");
 
@@ -524,7 +540,7 @@
 			// ResultType::FileLinked => todo!(),
 			(ResultType::BuildLogLine, [Str(s)]) => {
 				let s = ansi_filter(s);
-				info!("{s}");
+				info!(target: "nix::build", "{s}");
 			}
 			// ResultType::UntrustedPath => todo!(),
 			// ResultType::CorruptedPath => todo!(),
modifiedcrates/nix-eval/src/scheduler.rsdiffbeforeafterboth
--- a/crates/nix-eval/src/scheduler.rs
+++ b/crates/nix-eval/src/scheduler.rs
@@ -45,9 +45,36 @@
 	},
 }
 
+// Derivations whose names contain any of these are "heavy": memory-hungry builds that already
+// saturate all cores internally (e.g. composable-kernels), so running anything alongside them
+// just thrashes RAM. A heavy build grabs every semaphore permit and runs exclusively.
+const DEFAULT_HEAVY_DRVS: &[&str] = &[
+	"composable-kernel",
+	"composable_kernel",
+	"ghc",
+	"gfortran",
+	"llvm",
+	"clang",
+];
+
+fn heavy_patterns_from_env() -> Vec<String> {
+	let mut pats: Vec<String> = DEFAULT_HEAVY_DRVS.iter().map(|s| (*s).to_owned()).collect();
+	if let Ok(extra) = std::env::var("FLEET_HEAVY_DRVS") {
+		pats.extend(
+			extra
+				.split(',')
+				.map(str::trim)
+				.filter(|s| !s.is_empty())
+				.map(str::to_owned),
+		);
+	}
+	pats
+}
+
 pub struct Scheduler {
 	store: Arc<Store>,
 	parallelism: usize,
+	heavy_patterns: Vec<String>,
 	events: broadcast::Sender<BuildEvent>,
 }
 
@@ -58,10 +85,20 @@
 		Self {
 			store: eval_store(),
 			parallelism,
+			heavy_patterns: heavy_patterns_from_env(),
 			events,
 		}
 	}
 
+	// Permits a drv must hold to run. Heavy drvs take all of them, pausing everything else.
+	fn permits_for(&self, name: &str) -> u32 {
+		if self.heavy_patterns.iter().any(|p| name.contains(p)) {
+			self.parallelism as u32
+		} else {
+			1
+		}
+	}
+
 	pub fn subscribe(&self) -> broadcast::Receiver<BuildEvent> {
 		self.events.subscribe()
 	}
@@ -139,6 +176,7 @@
 						.get(&path)
 						.map(|n| n.name.clone())
 						.unwrap_or_default();
+					crate::logging::remove_drv_span(&path);
 					let _ = self.events.send(BuildEvent::DrvCancelled {
 						drv_path: path.clone(),
 						name,
@@ -153,8 +191,21 @@
 				let graph = graph.clone();
 				let wanted_here = wanted.get(&path).cloned().unwrap_or_default();
 				let store = self.store.clone();
+				let permits = graph
+					.nodes
+					.get(&path)
+					.map(|n| self.permits_for(&n.name))
+					.unwrap_or(1);
 				in_flight.push(tokio::spawn(async move {
-					let _permit = sem.acquire_owned().await.expect("semaphore not closed");
+					if permits > 1 {
+						debug!(permits, "heavy drv: waiting for exclusive build slot");
+					}
+					// Heavy drvs acquire every permit, so they only start once all in-flight
+					// builds drain and nothing new can slip in (the semaphore is FIFO-fair).
+					let _permit = sem
+						.acquire_many_owned(permits)
+						.await
+						.expect("semaphore not closed");
 					let node = graph
 						.nodes
 						.get(&path)
@@ -224,6 +275,7 @@
 					propagate_done(&dependents, &mut indeg, &mut ready, &finished);
 				}
 				Err(e) => {
+					crate::logging::remove_drv_span(&finished);
 					failed.insert(finished.clone(), format!("{e:#}"));
 					mark_tainted(&dependents, &finished, &mut tainted);
 					propagate_done(&dependents, &mut indeg, &mut ready, &finished);
@@ -368,11 +420,23 @@
 // TODO: Parallelism as a metric works poorly with multiple machines, but I haven't thought about bringing
 // hercy here yet. In case of remote machines - they will handle parallelism on their own, and this one
 // will work as a hard cap.
+/// Number of derivations to build concurrently, taken from Nix's `max-jobs`
+/// setting. `auto` resolves to the CPU count, matching Nix itself.
+fn max_jobs() -> usize {
+	let cores = || {
+		std::thread::available_parallelism()
+			.map(|p| p.get())
+			.unwrap_or(4)
+	};
+	match crate::get_setting(c"max-jobs") {
+		Ok(v) if v.trim() == "auto" => cores(),
+		Ok(v) => v.trim().parse().unwrap_or_else(|_| cores()),
+		Err(_) => cores(),
+	}
+}
+
 pub fn build_graph_sync(graph: Arc<DrvGraph>, root_outputs: Vec<String>) -> Result<()> {
-	let parallelism = std::thread::available_parallelism()
-		.map(|p| p.get())
-		.unwrap_or(4);
-	let scheduler = Scheduler::new(parallelism);
+	let scheduler = Scheduler::new(max_jobs());
 	crate::await_in_nix(async move { scheduler.run(graph, root_outputs).await })
 		.context("scheduler run")
 }
modifiedcrates/remowt-fleet/src/lib.rsdiffbeforeafterboth
--- a/crates/remowt-fleet/src/lib.rs
+++ b/crates/remowt-fleet/src/lib.rs
@@ -28,6 +28,8 @@
 	Sign(String),
 	#[error("listing generations failed: {0}")]
 	ListGenerations(String),
+	#[error("substitution failed: {0}")]
+	Substitute(String),
 }
 
 #[endpoints(ns = 91)]
@@ -60,6 +62,20 @@
 		.map_err(|e| NixError::Sign(e.to_string()))
 	}
 
+	/// Download the given paths (and their closures) into the local store using
+	/// the configured substituters. Used to fetch closures that were uploaded to
+	/// a binary cache (e.g. attic) instead of copied over the nix daemon tunnel.
+	#[endpoints(id = 6)]
+	async fn substitute(&self, paths: Vec<Utf8PathBuf>) -> Result<Vec<Utf8PathBuf>, NixError> {
+		spawn_blocking(move || {
+			let store = eval_store();
+			store.substitute_paths(&paths)
+		})
+		.await
+		.expect("substitution panicked")
+		.map_err(|e| NixError::Substitute(e.to_string()))
+	}
+
 	#[endpoints(id = 5)]
 	async fn list_generations(
 		&self,
addedmodules/cache.nixdiffbeforeafterboth
--- /dev/null
+++ b/modules/cache.nix
@@ -0,0 +1,61 @@
+{
+  lib,
+  fleetLib,
+  config,
+  ...
+}:
+let
+  inherit (lib.options) mkOption mkEnableOption literalExpression;
+  inherit (lib.types)
+    enum
+    str
+    bool
+    submodule
+    ;
+  inherit (fleetLib.options) mkHostsOption;
+
+  _file = ./cache.nix;
+
+  cacheOptions = {
+    enable = mkEnableOption "fleet binary cache";
+    kind = mkOption {
+      description = ''
+        Kind of the binary cache to use.
+      '';
+      type = enum [ "attic" ];
+      default = "attic";
+    };
+    name = mkOption {
+      description = ''
+        Name of the binary cache.
+      '';
+      type = str;
+    };
+    pin = mkOption {
+      description = ''
+        Whether to pin pushed paths in the attic cache, preventing them from
+        being garbage collected.
+
+        Requires compatible fork of attic server+client: https://github.com/zhaofengli/attic/pull/227#issuecomment-4759154796
+      '';
+      type = bool;
+      default = false;
+    };
+  };
+in
+{
+  options = {
+    cache = cacheOptions;
+    hosts = mkHostsOption {
+      inherit _file;
+      options.cache = mkOption {
+        description = ''
+          Binary cache configuration for this host.
+        '';
+        type = submodule { options = cacheOptions; };
+        default = config.cache;
+        defaultText = literalExpression "fleetConfiguration.cache";
+      };
+    };
+  };
+}
modifiedmodules/module-list.nixdiffbeforeafterboth
--- a/modules/module-list.nix
+++ b/modules/module-list.nix
@@ -1,5 +1,6 @@
 [
   ./assertions.nix
+  ./cache.nix
   ./fleetLib.nix
   ./hosts.nix
   ./nixos.nix
modifiedpkgs/default.nixdiffbeforeafterboth
--- a/pkgs/default.nix
+++ b/pkgs/default.nix
@@ -12,8 +12,6 @@
   fleet-generator-helper = callPackage ./fleet-generator-helper.nix { inherit craneLib; };
 
   inherit remowt-agents-bundle;
-  remowt-plugin-fleet = callPackage ./remowt-plugin-fleet.nix {
-    inherit craneLib inputs remowt-agents-bundle;
-  };
+  remowt-plugin-fleet = callPackage ./remowt-plugin-fleet.nix { inherit craneLib inputs; };
   remowt-ssh = callPackage ./remowt-ssh.nix { inherit craneLib remowt-agents-bundle; };
 }
modifiedremowt/cmds/remowt-agent/src/main.rsdiffbeforeafterboth
--- a/remowt/cmds/remowt-agent/src/main.rs
+++ b/remowt/cmds/remowt-agent/src/main.rs
@@ -273,10 +273,11 @@
 	let helper = SocketHelper {
 		fallback: SuidHelper,
 	};
-	register_auth_agent(&system_conn, Agent::new(helper, RofiPrompter)).await?;
+	let rofi = RofiPrompter::default();
+	register_auth_agent(&system_conn, Agent::new(helper, rofi.clone())).await?;
 
 	let mut rpc = Rpc::<BifConfig>::new(Address::User);
-	serve_prompts(&mut rpc, RofiPrompter);
+	serve_prompts(&mut rpc, rofi);
 
 	gateway::serve(rpc.clone(), &gateway::local_socket()?).await?;
 
modifiedremowt/cmds/remowt-ssh/Cargo.tomldiffbeforeafterboth
--- a/remowt/cmds/remowt-ssh/Cargo.toml
+++ b/remowt/cmds/remowt-ssh/Cargo.toml
@@ -8,6 +8,7 @@
 [dependencies]
 clap = { workspace = true, features = ["derive"] }
 tracing-subscriber.workspace = true
+tracing-journald.workspace = true
 remowt-link-shared.workspace = true
 remowt-client.workspace = true
 tokio = { workspace = true, features = [
modifiedremowt/cmds/remowt-ssh/src/main.rsdiffbeforeafterboth
--- a/remowt/cmds/remowt-ssh/src/main.rs
+++ b/remowt/cmds/remowt-ssh/src/main.rs
@@ -20,6 +20,8 @@
 use tokio::io::{AsyncRead, ReadBuf};
 use tokio::signal::unix::{signal, SignalKind};
 use tracing::debug;
+use tracing_subscriber::prelude::*;
+use tracing_subscriber::EnvFilter;
 
 #[derive(Parser)]
 enum Opts {
@@ -45,10 +47,16 @@
 
 #[tokio::main(flavor = "current_thread")]
 async fn main() -> anyhow::Result<()> {
-	tracing_subscriber::fmt()
-		.with_writer(std::io::stderr)
-		.without_time()
-		.init();
+	// Log to the journal, not stderr: this is an interactive client that puts
+	// the terminal in raw mode, so anything on stderr corrupts the session
+	// output. If the journal socket is unavailable, drop logs rather than fall
+	// back to stderr.
+	if let Ok(journald) = tracing_journald::layer() {
+		tracing_subscriber::registry()
+			.with(EnvFilter::from_default_env())
+			.with(journald.with_syslog_identifier("remowt-ssh".to_owned()))
+			.init();
+	}
 	let opts = Opts::parse();
 
 	let bundle = AgentBundle::from_dir(agents_dir()?)?;
modifiedremowt/crates/remowt-ui-prompt/Cargo.tomldiffbeforeafterboth
--- a/remowt/crates/remowt-ui-prompt/Cargo.toml
+++ b/remowt/crates/remowt-ui-prompt/Cargo.toml
@@ -12,5 +12,5 @@
 remowt-link-shared.workspace = true
 serde.workspace = true
 thiserror.workspace = true
-tokio = { workspace = true, features = ["io-util", "macros", "process", "rt"] }
+tokio = { workspace = true, features = ["io-util", "macros", "process", "rt", "sync"] }
 tracing.workspace = true
modifiedremowt/crates/remowt-ui-prompt/src/auto.rsdiffbeforeafterboth
--- a/remowt/crates/remowt-ui-prompt/src/auto.rs
+++ b/remowt/crates/remowt-ui-prompt/src/auto.rs
@@ -24,7 +24,7 @@
 		};
 		Self {
 			remote,
-			fallback: RofiPrompter,
+			fallback: RofiPrompter::default(),
 		}
 	}
 
modifiedremowt/crates/remowt-ui-prompt/src/rofi.rsdiffbeforeafterboth
--- a/remowt/crates/remowt-ui-prompt/src/rofi.rs
+++ b/remowt/crates/remowt-ui-prompt/src/rofi.rs
@@ -1,13 +1,18 @@
 use std::process::Stdio;
+use std::sync::Arc;
 
 use tokio::io::AsyncWriteExt;
 use tokio::process::Command;
+use tokio::sync::Mutex;
 use tracing::trace;
 
 use crate::{Error, Prompter, Result, Source};
 
-#[derive(Clone)]
-pub struct RofiPrompter;
+#[derive(Clone, Default)]
+pub struct RofiPrompter {
+	// Rofi can't run concurrently; serialize invocations.
+	lock: Arc<Mutex<()>>,
+}
 
 fn fixup_prompt(prompt: &str) -> &str {
 	// Rofi always appends such suffix
@@ -27,6 +32,7 @@
 		source: &[Source],
 	) -> Result<u32> {
 		trace!("rofi radio");
+		let _guard = self.lock.lock().await;
 		let mut cmd = rofi_command();
 		let mesg = if source.is_empty() {
 			description.to_owned()
@@ -113,6 +119,7 @@
 		source: &[Source],
 	) -> Result<String> {
 		trace!("rofi text");
+		let _guard = self.lock.lock().await;
 		let mut cmd = rofi_command();
 		let mesg = if source.is_empty() {
 			description.to_owned()
@@ -162,6 +169,7 @@
 
 	async fn display_text(&self, error: bool, description: &str, source: &[Source]) -> Result<()> {
 		trace!("rofi display");
+		let _guard = self.lock.lock().await;
 		let mut cmd = rofi_command();
 		let mut mesg = if source.is_empty() {
 			description.to_owned()
@@ -206,7 +214,7 @@
 	#[ignore = "interactive"]
 	async fn test() {
 		let prompter = PrependSourcePrompter {
-			prompter: RofiPrompter,
+			prompter: RofiPrompter::default(),
 			description: "test".to_owned(),
 			source: vec![Source(Cow::Borrowed("ssh"))],
 		};