trusty_console/

lib.rs

//! trusty-console library entry point.
//!
//! Why: Expose the console daemon's startup sequence as a public `run()`
//! function so bundled shim binaries inside host crates (trusty-search,
//! trusty-memory, trusty-analyze, trusty-review, trusty-mpm) can call
//! `trusty_console::run()` without duplicating any logic. This mirrors the
//! exact pattern used by trusty-embedderd (bundled into trusty-search via
//! issue #187) and trusty-bm25-daemon (bundled into trusty-memory via PR #190).
//! What: Re-exports all public submodules and provides `run_from(argv)` as the
//! canonical library entry point that parses an explicit argv vector and
//! dispatches to subcommands; `run()` is a thin wrapper passing the process's
//! global argv.
//! Test: `cargo test -p trusty-console` exercises the CLI parsing tests defined
//! in the submodules.

// docs.rs builds a release's documentation once, from the uploaded tarball,
// so a broken intra-doc link is baked into that version forever and only a new
// release can correct it. Deny keeps this crate at zero rather than letting the
// ratchet in `scripts/check_rustdoc_links.sh` absorb a new one.
#![deny(rustdoc::broken_intra_doc_links)]

use std::sync::Arc;
use std::time::Duration;

use anyhow::{Context, Result};
use clap::{Parser, Subcommand};
use tracing::info;
use trusty_common::{init_tracing, shutdown_signal, write_daemon_addr};

/// Default console HTTP bind address (used for both serve and `port` reporting).
///
/// Why: A single constant keeps the serve default and the `port` verb's
/// fallback in lock-step so `trusty-installer` (`tctl`) never discovers a port
/// the console would not actually bind to.
/// What: `127.0.0.1:7788` — the canonical localhost console address.
/// Test: `test_resolve_reported_addr_default` asserts the `port` verb falls
/// back to this value's host/port when no discovery file is present.
pub const DEFAULT_HTTP: &str = "127.0.0.1:7788";

/// Default console port, parsed once from [`DEFAULT_HTTP`].
///
/// Why: The `port` verb reports this when no running console has written a
/// discovery file yet.
/// What: `7788`.
/// Test: covered by the `port` verb tests below.
pub const DEFAULT_PORT: u16 = 7788;

pub mod bind;
// #6285: the console's own SPA, split out of `server` at the 500-SLOC cap.
pub mod connector;
pub mod console_ui;
pub mod detect;
// #6848: the console-hosted event bus core — UDS ingest, the bounded ring,
// dedup and eviction counters, and the subscriber fan-out a later slice
// (#6851) wires into an SSE route. No route surface in this crate yet.
pub(crate) mod event_bus;
// #6517: background whole-machine host-metrics sampler + cache feeding the
// machine-status route.
pub mod host_status;
// #6641: the bounded 10-minute sample window, the per-service transition log,
// and the SSE fan-out the Phase 3 graphs read.
pub mod machine_history;
pub mod mcp_handle;
pub mod metrics_poller;
pub mod poller;
pub mod proxy;
pub mod routes;
// #6285: the console's client to trusty-search, which no longer serves HTTP.
// #6155: the trusty-memory and trusty-analyze bridges, siblings of
// `search_uds`. Each maps one dashboard's REST-shaped paths onto that daemon's
// own `<domain>.*` methods.
pub mod analyze_uds;
pub mod memory_uds;
pub mod search_uds;
pub mod server;
pub mod service;
// #6642: per-service pid discovery + CPU sampling for the home-page graphs.
pub mod service_metrics;
// #9035: peer-identity gate for the `--tailscale` listener.
pub(crate) mod tailnet_peer;
// #6155: the trusty-search, trusty-memory and trusty-analyze SPAs, mounted
// under /tools/<tool>/.
pub mod tools_ui;
pub mod webhook;

/// How often the background sweep re-attempts pending webhook deliveries.
///
/// The sweep is the *recovery* mechanism, not the detection one — a stuck
/// delivery is detected by `GET /api/console/metrics/webhooks`, which scans the
/// spool on the request and therefore stays honest even if this loop dies.
const WEBHOOK_RETRY_INTERVAL: Duration = Duration::from_secs(60);
pub(crate) mod url_util;

// ─── CLI ─────────────────────────────────────────────────────────────────────

/// trusty-console: web dashboard for trusty services.
///
/// Why: Provides a single entry point for all console subcommands so future
/// phases (status, doctor, open) can be added without breaking existing usage.
/// What: Parses top-level arguments and delegates to subcommand handlers.
/// Test: `cargo run -p trusty-console -- serve --help` must succeed.
#[derive(Debug, Parser)]
#[command(
    name = "trusty-console",
    version,
    about = "Web dashboard for trusty services"
)]
pub struct Cli {
    #[command(subcommand)]
    pub command: Commands,
}

/// Available subcommands.
///
/// Why: `serve` runs the dashboard; `port` is a non-serving contract verb that
/// reports the console's bound (or default) port so orchestrators like
/// `trusty-installer` (`tctl`) can discover/launch the console without parsing logs.
/// What: Clap enum; each variant carries its own args.
/// Test: Subcommand selection tested via `Cli::parse_from`.
#[derive(Debug, Subcommand)]
pub enum Commands {
    /// Start the HTTP server and serve the console dashboard.
    Serve(ServeArgs),
    /// Report the console's bound (or default) HTTP port and exit.
    Port(PortArgs),
    /// Manage inference provider configuration (API keys) — the universal
    /// `config keys set/list/test/unset` surface shared by every trusty-*
    /// binary (epic #2400 Wave 1, #2405).
    Config(trusty_common::inference::config::ConfigCommand),
    /// Manage the macOS launchd LaunchAgent for the console daemon (#2557).
    ///
    /// `install` writes `~/Library/LaunchAgents/com.trusty.console.plist` (#8253)
    /// (running `trusty-console serve`) and bootstraps it; `uninstall` unloads
    /// and removes it; `status` / `logs` inspect the running agent. macOS-only.
    /// `tctl install` / `tctl start` call `install` on the operator's behalf.
    Service {
        #[command(subcommand)]
        action: service::ServiceAction,
    },
}

/// Arguments for `trusty-console port`.
///
/// Why: `trusty-installer` (`tctl`) (the orchestrator, issue #1316) discovers
/// the console URL by spawning `trusty-console port --json` and parsing the
/// `{addr,port}` envelope (see trusty-installer `os_env.rs`). The verb must
/// exist and be machine-readable for that discovery to work after the console is
/// de-bundled from the host crates (#1318).
/// What: A single `--json` flag selecting JSON output (default is a single
/// human-readable port line).
/// Test: `test_port_args_json_flag` / `test_port_args_default` below.
#[derive(Debug, Parser)]
pub struct PortArgs {
    /// Emit a JSON envelope `{"addr":"<host>","port":<u16>}` instead of a bare
    /// port number. Consumed by `trusty-installer` (`tctl`) console discovery.
    #[arg(long, default_value_t = false)]
    pub json: bool,
}

/// Arguments for `trusty-console serve`.
///
/// Why: The bind address must be configurable so users can change the port when
/// 7788 is taken; `--open` is a convenience for developers; `--poll-interval`
/// lets operators tune the background health-poll frequency; `--tailscale`
/// enables durable tailnet exposure without requiring `--http 0.0.0.0`.
/// What: Optional `--http` (default `127.0.0.1:7788`), `--open`,
/// `--poll-interval`, and `--tailscale` flags.
/// Env overrides: `TRUSTY_CONSOLE_BIND` sets the default bind mode so a
/// supervised/relaunched daemon stays tailnet-reachable without extra flags.
/// Test: Default address tested in `test_serve_args_defaults` below.
#[derive(Debug, Parser)]
pub struct ServeArgs {
    /// Address to listen on (default: 127.0.0.1:7788).
    ///
    /// Takes precedence over --tailscale and TRUSTY_CONSOLE_BIND when set to a
    /// non-default value.
    #[arg(long, default_value = "127.0.0.1:7788")]
    pub http: String,

    /// Expose the console on both 127.0.0.1 and the machine's Tailscale IPv4,
    /// enabling tailnet clients to reach the console without LAN exposure.
    ///
    /// The Tailscale IP is detected via `tailscale ip -4`. If Tailscale is not
    /// running, prints a warning and falls back to localhost-only.
    ///
    /// Can also be set persistently via the TRUSTY_CONSOLE_BIND=tailscale env
    /// var so a supervised/relaunched console stays tailnet-reachable without
    /// manually passing this flag.
    #[arg(long, default_value_t = false)]
    pub tailscale: bool,

    /// Open the console in the default browser after starting.
    #[arg(long, default_value_t = false)]
    pub open: bool,

    /// Background poll interval in seconds (default: 15).
    ///
    /// Controls BOTH the health poller (`poller::start`) AND the metrics
    /// poller (`metrics_poller::start`). Increasing this value reduces the
    /// frequency of both the HTTP health checks against each connector AND
    /// the stdio MCP `console_metrics` tool calls against trusty-analyze.
    #[arg(long, default_value_t = 15u64)]
    pub poll_interval: u64,

    /// Host-metrics sampling interval in seconds (default: 1).
    ///
    /// #6641: the cadence of the real-time graphs, kept separate from
    /// `--poll-interval` because they answer different questions. The service
    /// pollers each spawn a stdio-MCP round trip, so 15 s is deliberate there;
    /// a host sample is a local sysinfo refresh.
    ///
    /// #6642: the owner's home-page ruling is a bar per SECOND, so the default
    /// is 1 s and the window is 600 points. Raising this stretches the span the
    /// same 600 points cover, and the history payload advertises the value so
    /// the graph's x-axis follows.
    #[arg(
        long,
        default_value_t = trusty_common::host_metrics::history::HOST_SAMPLE_INTERVAL_SECS
    )]
    pub host_sample_interval: u64,

    /// Disk-metrics refresh interval in seconds (default: 15).
    ///
    /// The disk half of a host sample is the expensive half — on macOS
    /// `sysinfo` walks every mounted volume — so it runs on its own slower
    /// clock while CPU, memory and network keep `--host-sample-interval`.
    /// Free space changes on a human timescale; lower this only to watch a
    /// disk fill in near real time, and expect the CPU cost back.
    #[arg(
        long,
        default_value_t = trusty_common::host_metrics::history::DISK_SAMPLE_INTERVAL_SECS
    )]
    pub disk_sample_interval: u64,
}

// ─── public entry point ────────────────────────────────────────────────────

/// Library entry point for the trusty-console daemon, using the process argv.
///
/// Why: The standalone `trusty-console` crate is now the SOLE producer of the
/// `trusty-console` binary (#1318 — de-bundled from the 5 host crates). The
/// thin `main.rs` calls this, which simply forwards the process's global argv
/// to `run_from`.
/// What: Collects `std::env::args()` and delegates to [`run_from`].
/// Test: Indirectly via the `run_from` tests below and the binary smoke test.
pub async fn run() -> Result<()> {
    run_from(std::env::args().collect()).await
}

/// Library entry point parameterised on an explicit argv vector.
///
/// Why: Decoupling argument parsing from the process's global argv (#1318)
/// lets callers (tests, future embedders) drive the console deterministically
/// without mutating `std::env`. Previously `run()` called `Cli::parse()`,
/// which read global argv and could not be exercised in isolation.
/// What: Initialises tracing, parses `argv` via `Cli::parse_from`, and
/// dispatches to the matching subcommand handler. `argv[0]` is the program
/// name (clap convention). Returns `Ok(())` after clean shutdown.
/// Test: `test_run_from_port_json_outputs_envelope` drives this directly with
/// a synthetic argv; integration via `cargo test -p trusty-console`.
pub async fn run_from(argv: Vec<String>) -> Result<()> {
    init_tracing(1);

    let cli = Cli::parse_from(argv);

    match cli.command {
        Commands::Serve(args) => run_serve(args).await,
        Commands::Port(args) => run_port(args),
        Commands::Config(cmd) => cmd.run().await,
        // `service` drives macOS launchd synchronously; no async work needed.
        Commands::Service { action } => service::run_service_action(&action),
    }
}

/// Resolve the console's reportable HTTP address (host, port).
///
/// Why: The `port` verb must report the LIVE port of a running console when
/// one exists, falling back to the default otherwise — so `trusty-installer`
/// (`tctl`) discovery (issue #1316) points at the real dashboard, not a guess.
/// What: Reads the `trusty-console` discovery file via
/// `trusty_common::read_daemon_addr`; on a parseable `host:port` returns that
/// pair, else falls back to ([`DEFAULT_HTTP`] host, [`DEFAULT_PORT`]). Never
/// errors — discovery failures degrade to the default.
/// Test: `test_resolve_reported_addr_default` (no file → default).
pub fn resolve_reported_addr() -> (String, u16) {
    if let Ok(Some(recorded)) = trusty_common::read_daemon_addr("trusty-console")
        && let Ok(sa) = recorded.parse::<std::net::SocketAddr>()
    {
        return (sa.ip().to_string(), sa.port());
    }
    let default_host = DEFAULT_HTTP
        .rsplit_once(':')
        .map(|(h, _)| h.to_owned())
        .unwrap_or_else(|| "127.0.0.1".to_owned());
    (default_host, DEFAULT_PORT)
}

/// Run the `port` subcommand: print the console's bound/default port and exit.
///
/// Why: `trusty-installer` (`tctl`) console discovery spawns
/// `trusty-console port --json` and parses a `{addr,port}` envelope
/// (trusty-installer `os_env.rs`). This verb is the contract that makes that
/// discovery work; without it the call exits non-zero and console discovery is
/// silently broken (the latent bug fixed by #1318).
/// What: Resolves the reportable address; with `--json` prints
/// `{"addr":"<host>","port":<u16>}` to stdout, otherwise prints the bare port.
/// Returns `Ok(())`.
/// Test: `test_run_from_port_json_outputs_envelope`,
/// `test_port_envelope_is_valid_json`.
pub fn run_port(args: PortArgs) -> Result<()> {
    let (addr, port) = resolve_reported_addr();
    if args.json {
        let envelope = serde_json::json!({ "addr": addr, "port": port });
        println!("{envelope}");
    } else {
        println!("{port}");
    }
    Ok(())
}

/// Run the `serve` subcommand.
///
/// Why: Separating the serve logic from `run()` keeps `run()` thin and allows
/// this function to be called from integration tests.
/// What: Resolves bind addresses (respecting `--tailscale`, `--http`, and
/// `TRUSTY_CONSOLE_BIND`), builds the router, binds TCP listener(s), writes
/// the discovery file, starts the background health-poll task, optionally opens
/// a browser, then serves until SIGTERM/SIGINT with graceful shutdown.
/// Additional addresses beyond the primary get their own spawned `axum::serve`
/// task that runs concurrently until the shared shutdown signal fires.
/// Test: Server integration tests in `server.rs` cover the router directly
/// without exercising this function (to avoid real TCP binding in unit tests).
pub async fn run_serve(args: ServeArgs) -> Result<()> {
    // ── resolve bind mode ───────────────────────────────────────────────────
    let mode = bind::BindMode::from_env_and_flags(&args.http, DEFAULT_HTTP, args.tailscale);
    let port = bind::port_from_addr(&args.http, DEFAULT_PORT);
    let addrs = bind::resolve_bind_addrs(&mode, port, bind::detect_tailscale_ipv4);

    // ── service setup ───────────────────────────────────────────────────────
    let connectors = detect::all_connectors();
    let state = server::AppState::new(connectors);

    // Kick off an eager first poll so the cache is warm before the first
    // HTTP request arrives.
    {
        let cache = state.poller_cache().clone();
        let c = state.connectors();
        cache.poll_once(c).await;
    }

    // Start the background poller that refreshes the cache on the configured
    // interval.
    poller::start(
        state.poller_cache().clone(),
        state.connectors(),
        Duration::from_secs(args.poll_interval),
    );

    // ── metrics MCP poll (trusty-analyze) ───────────────────────────────────
    // The analyze handle is stored in AppState so on-demand routes
    // (/api/console/metrics/analyze/indexes, /api/console/metrics/analyze/visualize)
    // share the same child process. Here we hand a clone of that Arc to the
    // background metrics poller so both paths reuse one stdio connection.
    //
    // Why "mcp" not "serve --mcp":
    // `serve --mcp` starts BOTH the HTTP daemon and an MCP stdio loop; it
    // requires trusty-search to be reachable at startup and tries to open the
    // redb facts store (which may already be locked by the running daemon).
    // `mcp` only runs a pure stdio bridge pointing at the running HTTP daemon;
    // if the HTTP daemon is not yet up, `ensure_mcp_daemon_up` in analyze's
    // `mcp` subcommand starts it automatically. This is the correct invocation
    // for a lightweight stdio-only console_metrics child.
    metrics_poller::start(
        state.analyze_handle(),
        state.metrics_cache().clone(),
        Duration::from_secs(args.poll_interval),
    );

    // ── metrics MCP poll (trusty-memory) ────────────────────────────────────
    // trusty-memory's stdio MCP mode is `serve --stdio` (see main.rs).
    // The bridge forwards all JSON-RPC calls to the running HTTP daemon and
    // auto-starts it if absent. On machines without trusty-memory the handle
    // marks it Absent immediately; the cache stays None;
    // /api/console/metrics/memory returns 503 (graceful degradation).
    //
    // The handle comes from AppState::mcp_handles so the services route and
    // the metrics poller share the same McpServiceHandle (and thus the same
    // tools/list probe result — once the probe marks the handle Degraded,
    // that state is visible to both paths without a second probe).
    {
        let handles = state.mcp_handles();
        if let Some(h) = handles.get("trusty-memory") {
            metrics_poller::start(
                Arc::clone(h),
                state.memory_metrics_cache().clone(),
                Duration::from_secs(args.poll_interval),
            );
        } else {
            tracing::warn!(
                service = "trusty-memory",
                "run_serve: no MCP handle registered for trusty-memory — \
                 metrics poller will not start for this service"
            );
        }
    }

    // ── metrics MCP poll (trusty-search) ────────────────────────────────────
    // trusty-search's stdio MCP mode is `serve` (see serve_stdio in main.rs).
    // On machines without trusty-search the handle marks it Absent immediately;
    // the cache stays None; /api/console/metrics/search returns 503.
    //
    // Same shared-handle pattern as trusty-memory above.
    {
        let handles = state.mcp_handles();
        if let Some(h) = handles.get("trusty-search") {
            metrics_poller::start(
                Arc::clone(h),
                state.search_metrics_cache().clone(),
                Duration::from_secs(args.poll_interval),
            );
        } else {
            tracing::warn!(
                service = "trusty-search",
                "run_serve: no MCP handle registered for trusty-search — \
                 metrics poller will not start for this service"
            );
        }
    }

    // ── metrics MCP poll (trusty-review) ────────────────────────────────────
    // trusty-review's stdio MCP mode is `serve --stdio` (see commands/serve.rs).
    // When in stdio mode, trusty-review does NOT start an HTTP daemon — it runs
    // a pure MCP JSON-RPC loop over stdin/stdout, connected to the LLM directly.
    // This is the correct invocation for the console's lightweight metrics poll.
    // On machines without trusty-review the handle marks it Absent immediately;
    // the cache stays None; /api/console/metrics/review returns 503.
    //
    // Same shared-handle pattern as trusty-memory and trusty-search above.
    {
        let handles = state.mcp_handles();
        if let Some(h) = handles.get("trusty-review") {
            metrics_poller::start(
                Arc::clone(h),
                state.review_metrics_cache().clone(),
                Duration::from_secs(args.poll_interval),
            );
        } else {
            tracing::warn!(
                service = "trusty-review",
                "run_serve: no MCP handle registered for trusty-review — \
                 metrics poller will not start for this service"
            );
        }
    }

    // ── metrics MCP poll (trusty-mpm) ───────────────────────────────────────
    // trusty-mpm's stdio MCP mode is `serve --stdio` (the #1221 bridge that
    // auto-starts the durable daemon and forwards JSON-RPC to its loopback
    // POST /rpc). The console_metrics poll keeps the coarse session-fleet +
    // supervisor health cache warm for /api/console/metrics/mpm; the Sessions
    // tab itself polls /api/console/sessions live at a faster cadence (#1222).
    // On machines without trusty-mpm the handle marks it Absent immediately; the
    // cache stays None; /api/console/metrics/mpm returns 503 (graceful).
    {
        let handles = state.mcp_handles();
        if let Some(h) = handles.get("trusty-mpm") {
            metrics_poller::start(
                Arc::clone(h),
                state.mpm_metrics_cache().clone(),
                Duration::from_secs(args.poll_interval),
            );
        } else {
            tracing::warn!(
                service = "trusty-mpm",
                "run_serve: no MCP handle registered for trusty-mpm — \
                 metrics poller will not start for this service"
            );
        }
    }

    // ── whole-machine host sampler (#6517) + history feed (#6641) ───────────
    // One loop writes both the point-in-time cache
    // (GET /api/console/machine-status) and the bounded 10-minute window
    // (GET /api/console/machine-status/history and its SSE stream), so the two
    // can never disagree about the newest sample. It runs on its own 1 s
    // cadence rather than the service pollers' 15 s: a host sample is a local
    // sysinfo refresh, not an MCP round trip. #6642: the same loop also records
    // one per-service status + CPU sample per tick.
    machine_history::sampler::start_with_disk_interval(
        state.clone(),
        Duration::from_secs(args.host_sample_interval),
        Duration::from_secs(args.disk_sample_interval),
    );

    // #3269: trust the console's own non-loopback bind address(es) (e.g. the
    // Tailscale CGNAT address in `--tailscale` mode) as write-origin
    // self-origins, so the console's own write UI served from that address is
    // not 403'd by the same-origin guard. Loopback stays trusted unconditionally
    // regardless of bind mode.
    let self_origins = routes::origin_guard::SelfOrigins::from_bind_addrs(&addrs);

    // ── webhook ingress (#5089 step 3, ADR-0034) ────────────────────────────
    // `?` on purpose: a console that cannot open its spool must not start and
    // serve `/api/webhooks/{source}` anyway, because a delivery it cannot
    // durably record is a delivery it must refuse — and an unmounted route
    // would 404 instead of 5xx, which GitHub logs and no one reads.
    let ingress = webhook::WebhookIngress::from_env()
        .context("open the webhook spool under the console data directory")?;
    info!(
        spool = %ingress.spool().root().display(),
        "webhook ingress ready at POST /api/webhooks/{{source}}"
    );
    webhook::start_retry_sweep(ingress.clone(), WEBHOOK_RETRY_INTERVAL);
    let router = server::build_router_with_webhooks(state.clone(), self_origins, ingress);

    // ── event-bus ingest (#6848, DOC-73 §4.1/§4.2) ──────────────────────────
    // Console hosts the one event bus in the workspace; `trusty-mpm`,
    // `trusty-code`, `trusty-agents` and `trusty-analyze` push `HarnessEvent`
    // frames here over a UDS socket. Best-effort, unlike the webhook ingress
    // above: no route in this crate reads the bus yet (that is #6850/#6851),
    // and DOC-73 §4.1's "non-blocking invariant" makes the bus's own
    // availability an observability concern, never one console's HTTP surface
    // depends on — a console that could not bind this socket still serves
    // everything else.
    match event_bus::ingest_socket_path() {
        Ok(socket) => match event_bus::bind_ingest(&socket).await {
            Ok(listener) => {
                info!(
                    socket = %socket.display(),
                    "console event-bus ingest ready"
                );
                // #6848 slice 3b: open the durable NDJSON log before the bus
                // itself, so seq numbering resumes from the log's recovered
                // high-water mark rather than restarting at 1 on every
                // console restart (DOC-73 §4.3). Failing to open it degrades
                // to `EventBus::new` (no durability, seq restarts at 1) per
                // the same non-blocking invariant the ingest-bind failure
                // above already follows — a console that cannot open its
                // event log still serves everything else.
                let bus = match event_bus::LogConfig::resolve_default() {
                    Ok(log_config) => match event_bus::DurableLog::open(log_config).await {
                        Ok((log, recovered)) => {
                            info!(
                                next_seq = recovered.next_seq,
                                "console event-bus durable log ready"
                            );
                            event_bus::EventBus::with_log(
                                event_bus::EventBusConfig::default(),
                                Some(log),
                                recovered.next_seq,
                            )
                        }
                        Err(e) => {
                            tracing::warn!(
                                error = %e,
                                "could not open the console event-bus durable log; \
                                 events this run will not survive a restart"
                            );
                            event_bus::EventBus::new(event_bus::EventBusConfig::default())
                        }
                    },
                    Err(e) => {
                        tracing::warn!(
                            error = %e,
                            "could not resolve the console event-bus durable log directory; \
                             events this run will not survive a restart"
                        );
                        event_bus::EventBus::new(event_bus::EventBusConfig::default())
                    }
                };
                let bus = Arc::new(bus);
                tokio::spawn(async move {
                    event_bus::serve_ingest(listener, bus, shutdown_signal()).await;
                });
            }
            Err(e) => {
                tracing::warn!(
                    error = %e,
                    "could not bind the console event-bus ingest socket; \
                     producers cannot push events this run"
                );
            }
        },
        Err(e) => {
            tracing::warn!(error = %e, "could not resolve the console event-bus ingest socket path");
        }
    }

    // ── bind primary listener ───────────────────────────────────────────────
    let primary_addr = *addrs.first().context("bind address list is empty")?;
    let primary_listener = bind::bind_listener(primary_addr).await?;
    let primary_local = primary_listener.local_addr().context("get local addr")?;
    let addr_string = primary_local.to_string();
    info!("trusty-console listening on http://{primary_local}");

    // ── bind additional listeners (Tailscale mode: secondary addr) ──────────
    // #9035: each serves only nodes owned by this machine's own Tailscale
    // login, addressed to itself by exact Host/Origin; loopback is not gated.
    tailnet_peer::spawn_tailnet_listeners(
        addrs.get(1..).unwrap_or(&[]),
        &router,
        Arc::new(tailnet_peer::TailscaleCliResolver),
        shutdown_signal,
    )
    .await?;

    // ── write discovery file (primary address) ──────────────────────────────
    // Best-effort: log a warning on failure but do not abort the serve.
    if let Err(e) = write_daemon_addr("trusty-console", &addr_string) {
        tracing::warn!("could not write trusty-console discovery file: {e}");
    }

    let console_url = format!("http://{primary_local}");
    eprintln!("trusty-console: {console_url}");

    if args.open {
        // Best-effort browser open; ignore errors.
        let _ = open::that(&console_url);
    }

    axum::serve(primary_listener, router)
        .with_graceful_shutdown(shutdown_signal())
        .await
        .context("server error")?;

    // Best-effort removal of the discovery file on clean shutdown.
    // Only remove the file if it still points to our address; another
    // instance may have already written a new one.
    //
    // RESIDUAL RACE: the read → compare → delete sequence is not atomic. A
    // second instance could write a new address between our read and our
    // remove_file, causing us to delete a file we should not. The window is
    // tiny (milliseconds) and the consequence is cosmetic (a stale `port`
    // invocation returns the default rather than the live address). No
    // behavior change is required — this comment documents the known race.
    if let Ok(Some(recorded)) = trusty_common::read_daemon_addr("trusty-console")
        && recorded == addr_string
        && let Ok(dir) = trusty_common::resolve_data_dir("trusty-console")
    {
        let _ = std::fs::remove_file(dir.join("http_addr"));
    }

    Ok(())
}

// ─── tests ───────────────────────────────────────────────────────────────────

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

    // #6661: this module used to declare its own `TRUSTY_DATA_DIR_OVERRIDE`
    // mutex. Two locks over one process-global variable serialise nothing
    // between them, so a `remove_var` here landed inside the connector tests'
    // critical section. `detect::ENV_LOCK` is the crate's single lock.
    use crate::detect::ENV_LOCK as DATA_DIR_ENV_LOCK;

    /// Why: default http address must be 127.0.0.1:7788 and tailscale off.
    /// What: parses `serve` with no flags and checks all defaults.
    /// Test: this test itself.
    #[test]
    fn test_serve_args_defaults() {
        let cli = Cli::parse_from(["trusty-console", "serve"]);
        match cli.command {
            Commands::Serve(args) => {
                assert_eq!(args.http, "127.0.0.1:7788");
                assert!(!args.open);
                assert!(!args.tailscale);
                assert_eq!(args.poll_interval, 15);
                // #6642: the owner's home-page cadence — 1 s, distinct from the
                // 15 s service poll. Read from the constant so the default and
                // the window ruling cannot drift apart.
                assert_eq!(
                    args.host_sample_interval,
                    trusty_common::host_metrics::history::HOST_SAMPLE_INTERVAL_SECS
                );
                // The disk half keeps its own, slower default so the 1 s graph
                // does not pay for a volume walk every tick.
                assert_eq!(
                    args.disk_sample_interval,
                    trusty_common::host_metrics::history::DISK_SAMPLE_INTERVAL_SECS
                );
            }
            other => panic!("expected Serve, got {other:?}"),
        }
    }

    /// Why (#6641, #6642): the 1 s cadence is the shipped default, not a hard-coded
    /// constant — an operator who wants a longer window must be able to say so,
    /// and the two intervals must stay independent.
    /// What: parses `serve --host-sample-interval 10` and asserts only that
    /// value moved.
    /// Test: this test itself.
    #[test]
    fn test_serve_args_custom_host_sample_interval() {
        let cli = Cli::parse_from(["trusty-console", "serve", "--host-sample-interval", "10"]);
        match cli.command {
            Commands::Serve(args) => {
                assert_eq!(args.host_sample_interval, 10);
                assert_eq!(args.poll_interval, 15, "the service poll is unaffected");
                assert_eq!(
                    args.disk_sample_interval,
                    trusty_common::host_metrics::history::DISK_SAMPLE_INTERVAL_SECS,
                    "the disk cadence is unaffected"
                );
            }
            other => panic!("expected Serve, got {other:?}"),
        }
    }

    /// Why: --tailscale flag must be parsed correctly.
    /// What: parses `serve --tailscale`; asserts tailscale=true.
    /// Test: this test itself.
    #[test]
    fn test_serve_args_tailscale_flag() {
        let cli = Cli::parse_from(["trusty-console", "serve", "--tailscale"]);
        match cli.command {
            Commands::Serve(args) => {
                assert!(args.tailscale);
                assert_eq!(args.http, "127.0.0.1:7788");
            }
            other => panic!("expected Serve, got {other:?}"),
        }
    }

    /// Why: custom --http flag must override the default.
    /// What: parses `serve --http 0.0.0.0:9000`.
    /// Test: this test itself.
    #[test]
    fn test_serve_args_custom_http() {
        let cli = Cli::parse_from(["trusty-console", "serve", "--http", "0.0.0.0:9000"]);
        match cli.command {
            Commands::Serve(args) => {
                assert_eq!(args.http, "0.0.0.0:9000");
            }
            other => panic!("expected Serve, got {other:?}"),
        }
    }

    /// Why: --poll-interval must override the default.
    /// What: parses `serve --poll-interval 30`.
    /// Test: this test itself.
    #[test]
    fn test_serve_args_custom_poll_interval() {
        let cli = Cli::parse_from(["trusty-console", "serve", "--poll-interval", "30"]);
        match cli.command {
            Commands::Serve(args) => {
                assert_eq!(args.poll_interval, 30);
            }
            other => panic!("expected Serve, got {other:?}"),
        }
    }

    /// Why: the `port` subcommand must parse with a default (non-JSON) form so
    /// the bare-port output path is reachable.
    /// What: parses `port` and asserts `--json` defaults to false.
    /// Test: this test itself.
    #[test]
    fn test_port_args_default() {
        let cli = Cli::parse_from(["trusty-console", "port"]);
        match cli.command {
            Commands::Port(args) => assert!(!args.json),
            other => panic!("expected Port, got {other:?}"),
        }
    }

    /// Why: `trusty-installer` (`tctl`) invokes `trusty-console port --json`; the flag must parse.
    /// What: parses `port --json` and asserts `json == true`.
    /// Test: this test itself.
    #[test]
    fn test_port_args_json_flag() {
        let cli = Cli::parse_from(["trusty-console", "port", "--json"]);
        match cli.command {
            Commands::Port(args) => assert!(args.json),
            other => panic!("expected Port, got {other:?}"),
        }
    }

    /// Why: when no console has written a discovery file, the reported port
    /// must fall back to the canonical default so `trusty-installer` (`tctl`)
    /// still gets a usable address.
    /// What: calls `resolve_reported_addr` under an isolated data dir (no
    /// discovery file present) and asserts the default host/port.
    /// Test: this test itself; uses the data-dir override env to avoid reading
    /// a real running console's file.
    #[test]
    fn test_resolve_reported_addr_default() {
        let _guard = DATA_DIR_ENV_LOCK.lock().unwrap_or_else(|e| e.into_inner());
        let tmp = std::env::temp_dir().join(format!(
            "trusty-console-port-test-{}-{}",
            std::process::id(),
            std::time::SystemTime::now()
                .duration_since(std::time::UNIX_EPOCH)
                .map(|d| d.as_nanos())
                .unwrap_or(0)
        ));
        std::fs::create_dir_all(&tmp).expect("create temp data dir");
        // SAFETY: guarded by data_dir_test_lock to serialise env mutation.
        unsafe {
            std::env::set_var(trusty_common::DATA_DIR_OVERRIDE_ENV, &tmp);
        }
        let (addr, port) = resolve_reported_addr();
        unsafe {
            std::env::remove_var(trusty_common::DATA_DIR_OVERRIDE_ENV);
        }
        assert_eq!(addr, "127.0.0.1");
        assert_eq!(port, DEFAULT_PORT);
    }

    /// Why: the JSON envelope emitted by `run_port` must be valid JSON with the
    /// `addr` and `port` keys that `trusty-installer` (`tctl`) `parse_console_port` consumes.
    /// What: builds the same envelope `run_port` prints and round-trips it
    /// through serde to assert structure.
    /// Test: this test itself.
    #[test]
    fn test_port_envelope_is_valid_json() {
        let envelope = serde_json::json!({ "addr": "127.0.0.1", "port": DEFAULT_PORT });
        let s = envelope.to_string();
        let v: serde_json::Value = serde_json::from_str(&s).expect("valid json");
        assert_eq!(v.get("addr").and_then(|a| a.as_str()), Some("127.0.0.1"));
        assert_eq!(
            v.get("port").and_then(|p| p.as_u64()),
            Some(DEFAULT_PORT as u64)
        );
    }

    /// Why: the #1318 decoupling requires that an explicit argv parses to the
    /// `Port` command and that the `port` handler runs without touching the
    /// process's global argv. This exercises that parse → dispatch path.
    /// What: parses `["trusty-console","port","--json"]` via `Cli::parse_from`
    /// (the same call `run_from` makes) and runs `run_port` synchronously under
    /// an isolated data dir; asserts the dispatch matches `Port` and the
    /// handler returns Ok. Kept synchronous so the env-override mutex is never
    /// held across an `await` (clippy::await_holding_lock).
    /// Test: this test itself.
    #[test]
    fn test_run_from_port_json_outputs_envelope() {
        let _guard = DATA_DIR_ENV_LOCK.lock().unwrap_or_else(|e| e.into_inner());
        let tmp = std::env::temp_dir().join(format!(
            "trusty-console-runfrom-test-{}-{}",
            std::process::id(),
            std::time::SystemTime::now()
                .duration_since(std::time::UNIX_EPOCH)
                .map(|d| d.as_nanos())
                .unwrap_or(0)
        ));
        std::fs::create_dir_all(&tmp).expect("create temp data dir");
        // SAFETY: guarded by DATA_DIR_ENV_LOCK to serialise env mutation.
        unsafe {
            std::env::set_var(trusty_common::DATA_DIR_OVERRIDE_ENV, &tmp);
        }
        let argv = [
            "trusty-console".to_owned(),
            "port".to_owned(),
            "--json".to_owned(),
        ];
        let cli = Cli::parse_from(argv);
        let result = match cli.command {
            Commands::Port(args) => {
                assert!(args.json, "argv --json should parse to json=true");
                run_port(args)
            }
            other => panic!("expected Port, got {other:?}"),
        };
        unsafe {
            std::env::remove_var(trusty_common::DATA_DIR_OVERRIDE_ENV);
        }
        assert!(result.is_ok(), "run_port(port --json) should succeed");
    }

    /// Why: #8253 — the `service` clap doc comment (printed by `trusty-console
    /// service --help`) named a plist launchd never loaded, so an operator
    /// following it acted on a nonexistent unit.
    /// What: requires lib.rs's `Service` help text to name
    /// `<launchd_labels::CONSOLE>.plist` and never the pre-#4868 name.
    /// Test: pure string check on the embedded source, no fs side effects.
    #[test]
    fn service_help_names_the_registry_plist() {
        let src = include_str!("lib.rs");
        let want = format!(
            "~/Library/LaunchAgents/{}.plist",
            trusty_common::launchd_labels::CONSOLE
        );
        // Built piecewise so this test's own text never matches itself.
        let stale = format!("com.trusty.{}.plist", "trusty-console");
        assert!(src.contains(&want), "`service --help` must name {want}");
        assert!(
            !src.contains(&stale),
            "`service --help` names {stale}, a unit launchd never loaded"
        );
    }
}