podup 1.7.0

//! Intra-level parallelism for the lifecycle commands.
//!
//! The order resolver groups services into dependency *levels*
//! ([`crate::compose::resolve_levels`]): every service in one level has its
//! `depends_on` satisfied by an earlier level, so services *within* a level have
//! no ordering between them and can be acted on concurrently. The whole-project
//! lifecycle commands (stop/start/restart/kill/rm/pause/unpause) walk the levels
//! in order — preserving the cross-level dependency ordering — but dispatch each
//! level's per-service operations in parallel instead of strictly serially, so a
//! restart/stop of many independent services no longer serializes every grace
//! period (#757). This mirrors what the `up`/`create` path already does.

use std::collections::HashSet;

use crate::compose::types::{ComposeFile, Service};
use crate::engine::Engine;
use crate::error::{ComposeError, Result};
use crate::libpod::{urlencoded, API_PREFIX};

use super::targets::stop_timeout_param;

/// Upper bound on the number of same-level services a lifecycle command acts on
/// concurrently. Services within a dependency level have no ordering between
/// them, so they run in parallel; the cap keeps a very wide compose file from
/// opening an unbounded number of simultaneous libpod connections at once.
const MAX_LIFECYCLE_CONCURRENCY: usize = 16;

/// Run a batch of independent per-service futures concurrently, bounded by
/// [`MAX_LIFECYCLE_CONCURRENCY`], and return their results in the *input* order
/// (not completion order) so error selection and reporting stay deterministic
/// regardless of which service happens to finish first.
pub(super) async fn join_bounded<F>(futs: impl IntoIterator<Item = F>) -> Vec<Result<()>>
where
	F: std::future::Future<Output = Result<()>>,
{
	use futures_util::stream::StreamExt;
	let mut indexed: Vec<(usize, Result<()>)> = futures_util::stream::iter(
		futs.into_iter()
			.enumerate()
			.map(|(i, fut)| async move { (i, fut.await) }),
	)
	.buffer_unordered(MAX_LIFECYCLE_CONCURRENCY)
	.collect()
	.await;
	indexed.sort_by_key(|(i, _)| *i);
	indexed.into_iter().map(|(_, r)| r).collect()
}

/// Reduce a level's per-service results to the first error in service order, so
/// one failing service is still reported clearly while the rest of the level is
/// allowed to complete.
pub(super) fn first_error(results: Vec<Result<()>>) -> Option<ComposeError> {
	results.into_iter().find_map(Result::err)
}

/// Filter each dependency level down to `target_services`, dropping levels that
/// end up empty. An empty target list keeps every level. Returns an error if any
/// requested name is not in the file, matching [`super::targets::filter_services`]
/// (and docker compose's "no such service").
pub(super) fn filter_levels(
	file: &ComposeFile,
	levels: Vec<Vec<String>>,
	target_services: &[String],
) -> Result<Vec<Vec<String>>> {
	for name in target_services {
		if !file.services.contains_key(name) {
			return Err(ComposeError::ServiceNotFound(name.clone()));
		}
	}
	if target_services.is_empty() {
		return Ok(levels);
	}
	let set: HashSet<&str> = target_services.iter().map(|s| s.as_str()).collect();
	Ok(retain_levels(levels, |n| set.contains(n)))
}

/// Keep only the level entries matching `keep`, dropping levels left empty.
pub(super) fn retain_levels(
	levels: Vec<Vec<String>>,
	keep: impl Fn(&str) -> bool,
) -> Vec<Vec<String>> {
	levels
		.into_iter()
		.map(|level| {
			level
				.into_iter()
				.filter(|n| keep(n.as_str()))
				.collect::<Vec<_>>()
		})
		.filter(|level| !level.is_empty())
		.collect()
}

/// Compute the set of services a `restart` should act on, plus the explicit
/// target subset (used only to label cascade-restarts distinctly in the logs).
///
/// With no targets every service is restarted. With targets, the set is the
/// targets plus — unless `no_deps` — every service whose `depends_on` carries a
/// `restart: true` condition pointing at one of the targets (one hop, matching
/// the previous serial implementation).
pub(super) fn restart_service_set(
	file: &ComposeFile,
	target_services: &[String],
	no_deps: bool,
) -> (HashSet<String>, HashSet<String>) {
	if target_services.is_empty() {
		let all: HashSet<String> = file.services.keys().cloned().collect();
		return (all.clone(), all);
	}
	let targets: HashSet<String> = target_services.iter().cloned().collect();
	let mut full = targets.clone();
	if !no_deps {
		for (dep_name, dep_service) in &file.services {
			if targets
				.iter()
				.any(|t| dep_service.depends_on.restart_for(t))
			{
				full.insert(dep_name.clone());
			}
		}
	}
	(full, targets)
}

impl Engine {
	/// Stop a single service's live containers (only those actually running), as
	/// one unit of work in a concurrent level. See [`Engine::stop`].
	pub(super) async fn stop_one_service(&self, service_name: &str, grace: i32) -> Result<()> {
		for container in self.live_service_containers(service_name).await? {
			if super::scale::state_is_active(&container.state) {
				self.stop_container(&container.name, grace).await?;
			} else {
				tracing::debug!(
					"{}: not running ({}) — stop is a no-op",
					container.name,
					container.state
				);
			}
		}
		Ok(())
	}

	/// Start a single service's live containers, recording in `any_live` whether
	/// the service had any container to act on. See [`Engine::start`].
	pub(super) async fn start_one_service(
		&self,
		service_name: &str,
		any_live: &std::sync::atomic::AtomicBool,
	) -> Result<()> {
		let live = self
			.list_project_container_names(Some(service_name))
			.await?;
		if live.is_empty() {
			return Ok(());
		}
		any_live.store(true, std::sync::atomic::Ordering::Relaxed);
		let mut first_err: Option<ComposeError> = None;
		for container_name in live {
			let path = format!(
				"{API_PREFIX}/containers/{}/start",
				urlencoded(&container_name),
			);
			if let Err(e) = self
				.run_lifecycle_op(&path, &container_name, "Started")
				.await
			{
				first_err.get_or_insert(e);
			}
		}
		first_err.map_or(Ok(()), Err)
	}

	/// Restart a single service's live containers. `done` is the log verb
	/// (`restarted` for a direct target, `cascade-restarted` for a dependent).
	pub(super) async fn restart_one_service(
		&self,
		service_name: &str,
		service: &Service,
		done: &str,
	) -> Result<()> {
		let grace = self.grace_period_secs(service);
		let mut first_err: Option<ComposeError> = None;
		for container_name in self.live_replica_names(service_name, service).await? {
			// Single atomic restart (no visible stopped window) instead of a
			// stop+start round-trip.
			let restart_path = format!(
				"{API_PREFIX}/containers/{}/restart?t={}",
				urlencoded(&container_name),
				stop_timeout_param(grace),
			);
			if let Err(e) = self
				.run_lifecycle_op(&restart_path, &container_name, done)
				.await
			{
				first_err.get_or_insert(e);
			}
		}
		first_err.map_or(Ok(()), Err)
	}

	/// Send `signal` to a single service's live containers. See [`Engine::kill`].
	pub(super) async fn kill_one_service(
		&self,
		service_name: &str,
		service: &Service,
		signal: &str,
	) -> Result<()> {
		let mut first_err: Option<ComposeError> = None;
		for container_name in self.live_replica_names(service_name, service).await? {
			let path = format!(
				"{API_PREFIX}/containers/{}/kill?signal={}",
				urlencoded(&container_name),
				urlencoded(signal),
			);
			if let Err(e) = self
				.run_lifecycle_op(&path, &container_name, "Killed")
				.await
			{
				first_err.get_or_insert(e);
			}
		}
		first_err.map_or(Ok(()), Err)
	}

	/// Remove a single service's containers. See [`Engine::rm_with_options`].
	pub(super) async fn rm_one_service(
		&self,
		service_name: &str,
		service: &Service,
		force: bool,
		remove_volumes: bool,
	) -> Result<()> {
		let mut first_err: Option<ComposeError> = None;
		for container_name in self.live_replica_names(service_name, service).await? {
			let force_str = if force { "true" } else { "false" };
			let path = format!(
				"{API_PREFIX}/containers/{}?force={force_str}&v={remove_volumes}",
				urlencoded(&container_name),
			);
			match self.client.delete_existed(&path).await {
				// Only report a removal that actually happened — a phantom
				// (never-created) container 404s and must not be logged as
				// "removed".
				Ok(true) => crate::ui::progress_line("Container", &container_name, "Removed"),
				Ok(false) => {}
				// Without `--force`, a running container 409s. docker compose rm
				// skips running containers rather than aborting, so warn and keep
				// going (later stopped containers still get removed).
				Err(e) if !force && e.is_status(409) => {
					tracing::warn!(
						"{container_name} is running — skipping (pass -f to force removal)"
					);
				}
				Err(e) => {
					first_err.get_or_insert(ComposeError::Podman(e));
				}
			}
		}
		first_err.map_or(Ok(()), Err)
	}

	/// Pause or unpause a single service's live containers, treating a state
	/// mismatch as an idempotent no-op. `endpoint` is `pause`/`unpause`.
	pub(super) async fn idempotent_state_service(
		&self,
		service_name: &str,
		service: &Service,
		endpoint: &str,
		done: &str,
	) -> Result<()> {
		let mut first_err: Option<ComposeError> = None;
		for container_name in self.live_replica_names(service_name, service).await? {
			let path = format!(
				"{API_PREFIX}/containers/{}/{endpoint}",
				urlencoded(&container_name),
			);
			if let Err(e) = self
				.run_idempotent_state_op(&path, &container_name, done)
				.await
			{
				first_err.get_or_insert(e);
			}
		}
		first_err.map_or(Ok(()), Err)
	}
}

#[cfg(test)]
mod tests {
	use super::*;
	use crate::error::ComposeError;

	fn levels(input: &[&[&str]]) -> Vec<Vec<String>> {
		input
			.iter()
			.map(|lvl| lvl.iter().map(|s| s.to_string()).collect())
			.collect()
	}

	fn file_with(services: &[&str]) -> ComposeFile {
		let mut file = ComposeFile::default();
		for &s in services {
			file.services.insert(s.to_string(), Service::default());
		}
		file
	}

	#[tokio::test]
	async fn join_bounded_preserves_input_order() {
		// Futures finish out of order (later index resolves first) but the
		// collected results are returned in input order.
		let futs = (0..5usize).map(|i| async move {
			// Smaller i yields later via more yields, so completion order != input.
			for _ in 0..(5 - i) {
				tokio::task::yield_now().await;
			}
			if i == 2 {
				Err(ComposeError::ServiceNotFound(format!("svc{i}")))
			} else {
				Ok(())
			}
		});
		let results = join_bounded(futs).await;
		assert_eq!(results.len(), 5);
		// Only index 2 is an error, proving order is preserved.
		for (i, r) in results.iter().enumerate() {
			assert_eq!(r.is_err(), i == 2, "index {i}");
		}
	}

	#[test]
	fn first_error_returns_first_in_order() {
		let results = vec![
			Ok(()),
			Err(ComposeError::ServiceNotFound("a".into())),
			Err(ComposeError::ServiceNotFound("b".into())),
		];
		let err = first_error(results).unwrap();
		assert!(matches!(err, ComposeError::ServiceNotFound(n) if n == "a"));
	}

	#[test]
	fn first_error_none_when_all_ok() {
		assert!(first_error(vec![Ok(()), Ok(())]).is_none());
	}

	#[test]
	fn filter_levels_empty_targets_keeps_all() {
		let file = file_with(&["a", "b", "c"]);
		let lv = levels(&[&["a", "b"], &["c"]]);
		let out = filter_levels(&file, lv.clone(), &[]).unwrap();
		assert_eq!(out, lv);
	}

	#[test]
	fn filter_levels_drops_empty_levels() {
		let file = file_with(&["a", "b", "c"]);
		let lv = levels(&[&["a", "b"], &["c"]]);
		let out = filter_levels(&file, lv, &["c".into()]).unwrap();
		assert_eq!(out, levels(&[&["c"]]));
	}

	#[test]
	fn filter_levels_unknown_target_errors() {
		let file = file_with(&["a"]);
		let lv = levels(&[&["a"]]);
		let err = filter_levels(&file, lv, &["z".into()]).unwrap_err();
		assert!(matches!(err, ComposeError::ServiceNotFound(n) if n == "z"));
	}

	#[test]
	fn retain_levels_filters_and_drops() {
		let lv = levels(&[&["a", "b"], &["c"]]);
		let keep: HashSet<&str> = ["a", "c"].into_iter().collect();
		let out = retain_levels(lv, |n| keep.contains(n));
		assert_eq!(out, levels(&[&["a"], &["c"]]));
	}

	#[test]
	fn restart_service_set_empty_targets_is_all() {
		let file = file_with(&["a", "b"]);
		let (full, targets) = restart_service_set(&file, &[], false);
		assert_eq!(full, targets);
		assert_eq!(full.len(), 2);
		assert!(full.contains("a") && full.contains("b"));
	}

	#[test]
	fn restart_service_set_includes_cascade_dependents() {
		// web depends_on db with restart: true → restarting db cascades to web.
		let file = crate::parse_str(
			"services:\n  db:\n    image: x\n  web:\n    image: x\n    depends_on:\n      db:\n        condition: service_started\n        restart: true\n",
		)
		.unwrap();
		let (full, targets) = restart_service_set(&file, &["db".into()], false);
		assert!(targets.contains("db") && targets.len() == 1);
		assert!(full.contains("db") && full.contains("web"));
	}

	#[test]
	fn restart_service_set_no_deps_excludes_cascade() {
		let file = crate::parse_str(
			"services:\n  db:\n    image: x\n  web:\n    image: x\n    depends_on:\n      db:\n        condition: service_started\n        restart: true\n",
		)
		.unwrap();
		let (full, _) = restart_service_set(&file, &["db".into()], true);
		assert!(full.contains("db"));
		assert!(!full.contains("web"));
	}
}