hyperhive/hive-c0re/src/main.rs

use std::path::PathBuf;
use std::sync::Arc;

use anyhow::{Context as _, Result, bail};
use clap::{Parser, Subcommand};
use hive_sh4re::{HostRequest, HostResponse};

// Every module hangs off the `hive_c0re` library (see `src/lib.rs`).
// The daemon and the `hivectl` sibling binary share the same module
// tree — no per-binary duplication. Enumerated rather than wildcard
// so clippy stays happy + the lib surface this bin consumes is
// explicit (any new daemon entry point reads off the next add).
use hive_c0re::coordinator::{Coordinator, HiveEnv, ServeConfig};
use hive_c0re::{
    agent_sockets, auto_update, broker, client, crash_watch, dashboard, dashboard_events, forge,
    job_queue, knowledge, matrix, migrate, reminder_scheduler, scheduled_prompts_worker, server,
    socket_server,
};

#[derive(Parser)]
#[command(name = "hive-c0re", about = "hyperhive coordinator daemon and CLI")]
struct Cli {
    /// Path to the host admin socket.
    #[arg(long, global = true, default_value = "/run/hyperhive/host.sock")]
    socket: PathBuf,

    #[command(subcommand)]
    cmd: Cmd,
}

#[derive(Subcommand)]
enum Cmd {
    /// Run the coordinator daemon.
    Serve {
        /// Path to a JSON config file holding the host-level daemon config
        /// (the [`ServeConfig`](hive_c0re::coordinator::ServeConfig) shape:
        /// the container-injected `HiveEnv` fields + the hive-c0re-local
        /// `model_prices` table). Used as the base; any per-flag override
        /// below wins over the file. Absent → start from the built-in
        /// defaults. The NixOS module passes a generated config here so the
        /// `ExecStart` stays short instead of carrying every setting — and
        /// the context-window / price-table JSON blobs — as flags.
        #[arg(long)]
        config: Option<PathBuf>,
        /// Path to the sqlite message store.
        #[arg(long, default_value = hive_c0re::paths::BROKER_DB)]
        db: PathBuf,
        /// Override: URL of the hyperhive flake. Inlined into each
        /// per-agent `flake.nix` as the `hyperhive` input.
        #[arg(long)]
        hyperhive_flake: Option<String>,
        /// Override: store-path URL of the nixpkgs to wire into the meta
        /// flake as `inputs.nixpkgs.url`. Empty = legacy
        /// `follows = "hyperhive/nixpkgs"` fallback.
        #[arg(long)]
        nixpkgs_flake: Option<String>,
        /// Override: dashboard HTTP port.
        #[arg(long)]
        dashboard_port: Option<u16>,
        /// Override: operator pronouns (free text). Threaded into each
        /// container's harness via `HIVE_OPERATOR_PRONOUNS`.
        #[arg(long)]
        operator_pronouns: Option<String>,
        /// Override: per-model context-window sizes, as a JSON object
        /// mapping model-family short name to token count. Threaded into
        /// each container as `HIVE_CONTEXT_WINDOW_TOKENS_<KEY_UPPER>` env
        /// vars. Set via the `services.hive-c0re.contextWindowTokens`
        /// NixOS option.
        #[arg(long)]
        context_window_tokens: Option<String>,
        /// Override: systemd `CPUQuota=` applied to every agent container
        /// via a drop-in (e.g. `"200%"` = 2 cores). Set via
        /// `services.hyperhive.agentCpuQuota`.
        #[arg(long)]
        agent_cpu_quota: Option<String>,
        /// Override: systemd `MemoryMax=` applied to every agent
        /// container. Set via `services.hyperhive.agentMemoryMax`.
        #[arg(long)]
        agent_memory_max: Option<String>,
        /// Override: per-model USD prices (per million tokens) for the
        /// hive-wide ST4TS cost estimate, as a JSON object mapping a
        /// model-family short name to `{input, output, cache_read,
        /// cache_write}`. Models not covered fall back to the built-in
        /// estimate. Set via the `services.hyperhive.modelPrices` NixOS
        /// option.
        #[arg(long)]
        model_prices: Option<String>,
        /// Override: number of concurrent nix-heavy job-queue nodes
        /// (prebuild / profile-swap / create / meta lock). Set via the
        /// `services.hyperhive.c0re.buildSlots` NixOS option.
        #[arg(long)]
        build_slots: Option<usize>,
    },
    /// Spawn a new agent container directly (`hive-agent-<name>`). Bypasses
    /// the approval queue — use only as an operator on the host. For
    /// approval-gated spawns, use `request-spawn` instead.
    Spawn { name: String },
    /// Queue a spawn request as an approval. The container is created on
    /// `approve <id>` (CLI) or the dashboard's APPR0VE button.
    RequestSpawn { name: String },
    /// Stop a managed container (graceful).
    Kill { name: String },
    /// Tear down a sub-agent container. Container is removed; persistent
    /// state (config repos + Claude credentials) is kept by default. Pass
    /// `--purge` to also wipe the agent's state dirs (config + creds +
    /// notes). No undo.
    Destroy {
        name: String,
        #[arg(long)]
        purge: bool,
    },
    /// Apply pending config to a managed container.
    Rebuild { name: String },
    /// List managed containers.
    List,
    /// List pending approval requests submitted by the manager.
    Pending,
    /// Approve a pending request by id; the action runs immediately.
    Approve { id: i64 },
    /// Deny a pending request by id.
    Deny { id: i64 },
    /// Move an agent in the topology tree. Set `--parent` to a new
    /// parent agent name; pass `--root` to promote the agent to root
    /// (no parent). Refuses cycles and unknown agents. The manager
    /// is reparentable like any other agent — its privileges come
    /// from the privileged MCP socket, not its tree position.
    SetParent {
        child: String,
        /// New parent agent name. Mutually exclusive with `--root`.
        /// Exactly one of `--parent` / `--root` is required — clap
        /// rejects both-absent calls so a fat-fingered
        /// `hive-c0re set-parent alice` doesn't silently promote
        /// alice to root.
        #[arg(long, conflicts_with = "root", required_unless_present = "root")]
        parent: Option<String>,
        /// Promote `child` to root (no parent).
        #[arg(long)]
        root: bool,
    },
}

#[tokio::main]
async fn main() -> Result<()> {
    tracing_subscriber::fmt()
        .with_env_filter(
            tracing_subscriber::EnvFilter::try_from_default_env()
                .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("info")),
        )
        .init();

    let cli = Cli::parse();
    match cli.cmd {
        Cmd::Serve {
            config,
            db,
            hyperhive_flake,
            nixpkgs_flake,
            dashboard_port,
            operator_pronouns,
            context_window_tokens,
            agent_cpu_quota,
            agent_memory_max,
            model_prices,
            build_slots,
        } => {
            // Base config from the --config file (or the built-in
            // defaults), then apply any per-flag overrides — config
            // file is the base, explicit flags win.
            let mut sc = match &config {
                Some(p) => {
                    let s = std::fs::read_to_string(p)
                        .with_context(|| format!("read --config {}", p.display()))?;
                    serde_json::from_str::<ServeConfig>(&s)
                        .with_context(|| format!("parse --config {}", p.display()))?
                }
                None => ServeConfig::default(),
            };
            if let Some(v) = hyperhive_flake {
                sc.env.hyperhive_flake = v;
            }
            if let Some(v) = nixpkgs_flake {
                sc.env.nixpkgs_flake = v;
            }
            if let Some(v) = dashboard_port {
                sc.env.dashboard_port = v;
            }
            if let Some(v) = operator_pronouns {
                sc.env.operator_pronouns = v;
            }
            if let Some(v) = context_window_tokens {
                sc.env.context_window_tokens =
                    serde_json::from_str(&v).context("--context-window-tokens: invalid JSON")?;
            }
            if let Some(v) = agent_cpu_quota {
                sc.env.agent_cpu_quota = v;
            }
            if let Some(v) = agent_memory_max {
                sc.env.agent_memory_max = v;
            }
            if let Some(v) = model_prices {
                sc.model_prices =
                    serde_json::from_str(&v).context("--model-prices: invalid JSON")?;
            }
            if let Some(v) = build_slots {
                sc.build_slots = v;
            }
            cmd_serve(sc.env, sc.model_prices, sc.build_slots, db, &cli.socket).await
        }
        Cmd::Spawn { name } => {
            render(client::request(&cli.socket, HostRequest::Spawn { name }).await?)
        }
        Cmd::RequestSpawn { name } => {
            render(client::request(&cli.socket, HostRequest::RequestSpawn { name }).await?)
        }
        Cmd::Kill { name } => {
            render(client::request(&cli.socket, HostRequest::Kill { name }).await?)
        }
        Cmd::Destroy { name, purge } => {
            render(client::request(&cli.socket, HostRequest::Destroy { name, purge }).await?)
        }
        Cmd::Rebuild { name } => {
            render(client::request(&cli.socket, HostRequest::Rebuild { name }).await?)
        }
        Cmd::List => render(client::request(&cli.socket, HostRequest::List).await?),
        Cmd::Pending => render(client::request(&cli.socket, HostRequest::Pending).await?),
        Cmd::Approve { id } => {
            render(client::request(&cli.socket, HostRequest::Approve { id }).await?)
        }
        Cmd::Deny { id } => render(client::request(&cli.socket, HostRequest::Deny { id }).await?),
        Cmd::SetParent {
            child,
            parent,
            root,
        } => {
            let new_parent = if root { None } else { parent };
            render(
                client::request(&cli.socket, HostRequest::SetParent { child, new_parent }).await?,
            )
        }
    }
}

/// Start the coordinator daemon: open the broker, run migrations, spawn
/// background tasks (auto-update, vacuums, crash-watcher, reminder-scheduler,
/// dashboard), then serve the admin socket until a signal arrives.
#[allow(
    clippy::too_many_lines,
    reason = "startup orchestration: open the broker, run migrations, then spawn \
              the full set of background services (auto-update, vacuums, \
              crash-watch, schedulers, dashboard) before the serve loop — the \
              length is inherent to booting the daemon, not the arg list"
)]
async fn cmd_serve(
    env: HiveEnv,
    model_prices: hive_c0re::hive_stats::PriceTable,
    build_slots: usize,
    db: std::path::PathBuf,
    socket: &std::path::Path,
) -> Result<()> {
    // Move any host-side state still at the legacy flat layout into its
    // subdir (`db/`, `forge/`, `matrix/`, `run/`) BEFORE opening the
    // broker db — the broker + build-logs dbs are among the relocated
    // files. Idempotent; a no-op once migrated.
    hive_c0re::paths::relocate_legacy_state();
    // `dashboard_port` is consumed into the Coordinator below; capture the
    // Copy value first for the dashboard + knowledge-webhook tasks.
    let dashboard_port = env.dashboard_port;
    let coord = Arc::new(Coordinator::open(&db, env, model_prices, build_slots)?);
    socket_server::start_manager(coord.clone())?;
    // Idempotent pre-flight: rewrite pre-meta-layout applied
    // repos, ensure proposed repos carry the `applied`
    // remote, bootstrap the meta repo, repoint containers at
    // `meta#<name>` (one-shot, guarded by a marker file).
    // Runs before manager auto-spawn so the new manager is
    // built against meta from the first attempt.
    if let Err(e) = migrate::run(&coord).await {
        tracing::warn!(error = ?e, "startup migration failed");
    }
    // Auto-create the root agent container if it isn't there yet. Block
    // on this — without root the system has no manager harness.
    // Failures are logged but allowed: a broken auto-spawn shouldn't
    // make the dashboard unreachable for debugging.
    if let Err(e) = auto_update::ensure_root_agent(&coord).await {
        tracing::warn!(error = ?e, "auto-spawn root agent failed");
    }
    // Sync /etc/tmpfiles.d/hyperhive-agents.conf so agent runtime dirs are
    // pre-declared for the next boot. Best-effort background task — a failure
    // here must not block hive-c0re startup. See lifecycle::sync_tmpfiles.
    tokio::spawn(async {
        hive_c0re::lifecycle::sync_tmpfiles().await;
    });
    // Auto-update in the background — don't block service start.
    // Sub-agent rebuilds can take tens of seconds; we want the admin
    // socket up immediately.
    let update_coord = coord.clone();
    tokio::spawn(async move {
        if let Err(e) = auto_update::run(update_coord).await {
            tracing::warn!(error = ?e, "auto-update task failed");
        }
    });
    // Forge user sweep: ensure every existing container has a
    // forgejo user + access token. No-op when the hive-forge
    // container isn't running. Backgrounded — touches the
    // forge state dir via `nixos-container run` which is slow.
    tokio::spawn(async move {
        forge::ensure_all().await;
    });
    // Knowledge webhook setup: ensure the Forgejo push webhook for
    // `internal/knowledge` exists so `pull()` fires on merge. Runs
    // after forge::ensure_all so the core token + repo are present.
    // No-op when the core token or forge are absent.
    let webhook_port = dashboard_port;
    tokio::spawn(async move {
        if let Some(token) = forge::core_token()
            && let Err(e) = knowledge::ensure_webhook(&token, webhook_port).await
        {
            tracing::warn!(error = ?e, "knowledge: ensure_webhook failed");
        }
    });
    // Knowledge periodic pull: hourly fallback in case the webhook is
    // missed (e.g. hive-c0re was down during a push). First fires at
    // startup (immediate pull after the clone is already present).
    let mut knowledge_shutdown = coord.shutdown_rx();
    tokio::spawn(async move {
        // Initial pull — reconcile any commits that landed while c0re
        // was offline.
        if let Err(e) = knowledge::pull().await {
            tracing::debug!(error = ?e, "knowledge: startup pull skipped (no clone yet?)");
        }
        let interval = std::time::Duration::from_hours(1);
        loop {
            tokio::select! {
                () = tokio::time::sleep(interval) => {
                    if let Err(e) = knowledge::pull().await {
                        tracing::warn!(error = ?e, "knowledge: periodic pull failed");
                    }
                }
                _ = knowledge_shutdown.changed() => {
                    tracing::info!("knowledge pull: shutdown signal received");
                    break;
                }
            }
        }
    });
    // Matrix user sweep: same shape — ensure every container has
    // an account on the local matrix-tuwunel homeserver with an
    // access_token persisted to `<state>/matrix-token`. No-op when
    // the hive-matrix container isn't running. Backgrounded because
    // UIAA is a two-roundtrip dance per agent.
    //
    // Runs once at startup AND periodically every 30 minutes so that
    // token files deleted by `hive-matrix-daemon` (stale-token
    // recovery — `M_UNKNOWN_TOKEN`) get re-provisioned without
    // requiring a hive-c0re restart.
    let mut matrix_shutdown = coord.shutdown_rx();
    tokio::spawn(async move {
        let interval = std::time::Duration::from_mins(30);
        matrix::ensure_all().await;
        loop {
            tokio::select! {
                () = tokio::time::sleep(interval) => {
                    matrix::ensure_all().await;
                }
                _ = matrix_shutdown.changed() => {
                    tracing::info!("matrix ensure_all: shutdown signal received");
                    break;
                }
            }
        }
    });
    // Periodic broker vacuum: drop fully-acked messages older
    // than 30 days. Delivered-but-unacked rows (recoverable via
    // requeue_inflight) and undelivered rows are always kept.
    // Runs hourly; first sweep happens immediately.
    let vacuum_coord = coord.clone();
    let mut vacuum_shutdown = coord.shutdown_rx();
    tokio::spawn(async move {
        let interval = std::time::Duration::from_hours(1);
        let keep_secs: i64 = 30 * 24 * 3600;
        loop {
            match vacuum_coord.broker.vacuum_delivered(keep_secs) {
                Ok(0) => {}
                Ok(n) => tracing::info!(removed = n, "broker vacuum"),
                Err(e) => tracing::warn!(error = ?e, "broker vacuum failed"),
            }
            tokio::select! {
                () = tokio::time::sleep(interval) => {}
                _ = vacuum_shutdown.changed() => {
                    tracing::info!("broker vacuum: shutdown signal received");
                    break;
                }
            }
        }
    });
    // Per-agent events.sqlite + bash-tasks file cleanup now runs
    // agent-side in the harness (`hive_ag3nt::vacuum`): the files are
    // agent-owned, so host-side deletes hit PermissionDenied / readonly-db
    // under privsep. See issue tracker "perms borked".
    // (turn-stats.sqlite has no vacuum — it's one tiny row per turn,
    // ~hundreds of KB, and the /stats + hive-stats views read it
    // directly; pruning it would just lose trend history for no gain.)
    // Slow per-container disk sampler: a `du` of each agent's state dir
    // + writable rootfs every ~5 min, cached so the 5s container-load
    // poll stays cheap cgroup-only reads. Feeds `disk_bytes` on the LOAD
    // tab. See container_stats::disk_sampler_loop.
    hive_c0re::container_stats::spawn_disk_sampler();
    // build_logs.sqlite vacuum: c0re-side (single db). Failures kept
    // 30d, successes 24h — see `build_logs::vacuum` for the rule.
    hive_c0re::build_logs::spawn_vacuum(&coord);
    // audit_log.sqlite vacuum: agent-initiated privileged-action trail,
    // 90d retention — see `audit_log::vacuum`.
    hive_c0re::audit_log::spawn_vacuum(&coord);
    // Container crash watcher: emits HelperEvent::ContainerCrash
    // when a previously-running container goes away without an
    // operator-initiated transient state.
    crash_watch::spawn(coord.clone());
    // Agent-sockets marker poll: re-fires `agent_sockets::write`
    // and `gateway_nginx::write` every 10s so the JSON and nginx
    // config pick up newly-bound `.bound` markers after a rebuild.
    // Also retries any pending gateway nginx reload that failed on
    // the previous tick. write() is idempotent so steady-state cost
    // is one stat per agent per tick.
    // See `docs/gateway.md::Per-agent unix-socket upstream`.
    agent_sockets::spawn_poll();
    // Reminder scheduler: drains due reminders + handles
    // file_path payload persistence. See reminder_scheduler.rs.
    reminder_scheduler::spawn(coord.clone());
    // Scheduled-prompts worker: drains due scheduled_prompts rows
    // and fans the body out to each active target's inbox. See
    // scheduled_prompts_worker.rs.
    scheduled_prompts_worker::spawn(coord.clone());
    // Job-queue scheduler: drives the global DAG queue (rebuild /
    // meta-update / spawn / power ops). Concurrency comes from the
    // build-slot count + per-agent leases inside the queue, not from
    // multiple workers — cheap nodes (graceful signals, drains,
    // reconciles) overlap nix-heavy ones structurally. Call sites
    // (auto_update, dashboard, manager, approval handler) submit DAGs
    // instead of awaiting lifecycle work inline. See `job_queue/`.
    {
        let q_coord = coord.clone();
        tokio::spawn(async move {
            job_queue::scheduler::run_worker(q_coord).await;
        });
    }
    // Forward every broker event onto the unified dashboard
    // channel with a freshly-stamped seq, so the dashboard SSE
    // sees broker messages + future mutation events on one
    // stream with one monotonic seq. The broker's intra-process
    // channel (used by `recv_blocking_batch`) stays untouched.
    spawn_broker_to_dashboard_forwarder(coord.clone());
    let dash_coord = coord.clone();
    tokio::spawn(async move {
        if let Err(e) = dashboard::serve(dashboard_port, dash_coord).await {
            tracing::error!(error = ?e, "dashboard failed");
        }
    });
    // Run the admin socket until a signal arrives; then signal
    // all background tasks so they exit cleanly before the
    // process terminates.
    let coord_sig = coord.clone();
    tokio::select! {
        res = server::serve(socket, coord) => { res? }
        _ = tokio::signal::ctrl_c() => {
            tracing::info!("SIGINT received — requesting shutdown");
            coord_sig.request_shutdown();
        }
        () = async {
            let mut sig = tokio::signal::unix::signal(
                tokio::signal::unix::SignalKind::terminate()
            ).expect("failed to install SIGTERM handler");
            sig.recv().await;
        } => {
            tracing::info!("SIGTERM received — requesting shutdown");
            coord_sig.request_shutdown();
        }
    }
    Ok(())
}

/// Re-emit every broker `MessageEvent` onto the dashboard channel as
/// a `DashboardEvent::Sent` / `Delivered` with a freshly-stamped seq.
/// Background task; runs for the life of the process. On a lagged
/// broker subscription we just keep going — the dashboard channel is
/// best-effort presentation plumbing, the broker keeps its own sqlite
/// log for replay.
fn spawn_broker_to_dashboard_forwarder(coord: Arc<Coordinator>) {
    use broker::MessageEvent;
    use dashboard_events::DashboardEvent;
    let mut rx = coord.broker.subscribe();
    tokio::spawn(async move {
        loop {
            match rx.recv().await {
                Ok(MessageEvent::Sent {
                    id,
                    from,
                    to,
                    body,
                    at,
                    in_reply_to,
                }) => {
                    let file_refs = dashboard::scan_validated_paths(&body);
                    coord.emit_dashboard_event(DashboardEvent::Sent {
                        seq: coord.next_seq(),
                        id,
                        from,
                        to,
                        body,
                        at: hive_sh4re::wire_time::from_secs(at),
                        in_reply_to,
                        file_refs,
                    });
                }
                Ok(MessageEvent::Delivered {
                    id,
                    from,
                    to,
                    body,
                    at,
                    in_reply_to,
                }) => {
                    let file_refs = dashboard::scan_validated_paths(&body);
                    coord.emit_dashboard_event(DashboardEvent::Delivered {
                        seq: coord.next_seq(),
                        id,
                        from,
                        to,
                        body,
                        at: hive_sh4re::wire_time::from_secs(at),
                        in_reply_to,
                        file_refs,
                    });
                }
                Err(tokio::sync::broadcast::error::RecvError::Lagged(n)) => {
                    tracing::warn!(skipped = n, "broker-to-dashboard forwarder lagged");
                }
                Err(tokio::sync::broadcast::error::RecvError::Closed) => break,
            }
        }
    });
}

fn render(resp: HostResponse) -> Result<()> {
    println!("{}", serde_json::to_string_pretty(&resp)?);
    if !resp.ok {
        bail!(resp.error.unwrap_or_else(|| "request failed".to_owned()));
    }
    Ok(())
}