//! Background health and SSH-banner check tasks, including the transition gate,
//! incident bookkeeping, and latency-drift alerting.

use tokio::task::JoinHandle;
use tracing::info;

use pom::alerts::Alerter;
use pom::checks::{drift, http, ssh_banner};
use pom::config::Config;
use pom::db;
use pom::types::{HealthStatus, LatencyStats};

use super::super::incident::{IncidentAction, incident_action};
use super::super::transition::TransitionGate;
use super::{CheckInterval, configured_targets};

pub(crate) fn spawn_health_tasks(
    config: &Config,
    pool: &sqlx::SqlitePool,
    cancel: &tokio_util::sync::CancellationToken,
    alerter: Option<&Alerter>,
) -> Vec<JoinHandle<()>> {
    let default_interval = config.serve.interval_secs;
    let confirmations = config.serve.confirmations;
    let mut handles = Vec::new();

    for (name, target_config) in configured_targets(config) {
        if let Some(health_config) = target_config.health {
            let interval_secs = health_config.interval_secs.unwrap_or(default_interval);
            let pool = pool.clone();
            let name = name.clone();
            let label = target_config.label.clone();
            let alerter = alerter.cloned();
            let cancel = cancel.clone();
            let trending_config = health_config.trending.clone();

            info!("{name}: health check every {interval_secs}s");

            handles.push(tokio::spawn(async move {
                let mut ticks = CheckInterval::new(interval_secs, cancel);
                let expect = health_config.expect.as_ref();
                let mut in_drift = false;
                // Seed the N-of-M transition gate from the last stored status so a
                // restart of a persistently-degraded target doesn't re-fire, and a
                // transient blip is debounced before it pages/opens an incident.
                let seed = db::get_latest_health(&pool, &name)
                    .await
                    .ok()
                    .flatten()
                    .map(|s| s.status);
                let mut gate = TransitionGate::seeded(seed, confirmations);

                // RECONCILE THE INCIDENT LEDGER AGAINST THE SEED, before the
                // first tick.
                //
                // Seeding is what stops a restart re-firing a transition, and it
                // is also how a restart can LOSE one. An incident closes on a
                // confirmed transition back to Operational; if the process
                // restarts across that recovery, the gate seeds Operational,
                // observes Operational, reports no transition, and the incident
                // stays open forever with nothing to reopen the question.
                //
                // Measured on astra 2026-08-23: incident 307, `mt`, opened
                // 2026-08-16 16:57 with `ended_at` still NULL a week later. Two
                // sibling incidents opened in the same minute by the same blip
                // (`mnw` 308, `htpy` 309) both closed an hour later, and mt's
                // health had been Operational at 100% uptime throughout. The
                // instance had read `failed` ever since, which is a monitor
                // reporting its own bookkeeping as an outage.
                //
                // Idempotent and once per target per start, so it costs one
                // query at boot and nothing afterwards. Deliberately does NOT
                // alert: this closes a stale record, and a recovery notice for
                // an outage that ended a week ago is noise.
                if seed == Some(HealthStatus::Operational) {
                    match db::close_open_incidents(&pool, &name).await {
                        Ok(0) => {}
                        Ok(n) => info!(
                            "{name}: closed {n} incident(s) left open across a restart;                              health reads operational"
                        ),
                        Err(e) => {
                            tracing::error!("{name}: failed to reconcile open incidents: {e}");
                        }
                    }
                }
                while ticks.next().await {
                    let snapshot = http::check_health(&name, &health_config, expect).await;
                    info!(
                        "{}: {} ({}ms)",
                        name, snapshot.status, snapshot.response_time_ms
                    );
                    if let Err(e) = db::insert_health_check(&pool, &snapshot).await {
                        tracing::error!("{name}: failed to store health check: {e}");
                    }

                    // Alert AND open/close incidents only on a *confirmed*
                    // transition (debounced + restart-seeded above).
                    if let Some((from_status, to_status)) = gate.observe(snapshot.status) {
                        if let Some(ref alerter) = alerter {
                            let from = from_status.to_string();
                            let to = to_status.to_string();
                            if to_status == HealthStatus::Operational {
                                alerter.send_health_recovery(&name, &label, &from).await;
                            } else {
                                alerter
                                    .send_health_alert(
                                        &name,
                                        &label,
                                        &from,
                                        &to,
                                        snapshot.error.as_deref(),
                                    )
                                    .await;
                            }
                        }

                        match incident_action(from_status, to_status) {
                            IncidentAction::None => {}
                            IncidentAction::Open => {
                                if let Err(e) = db::insert_incident(
                                    &pool,
                                    &name,
                                    &from_status.to_string(),
                                    &to_status.to_string(),
                                )
                                .await
                                {
                                    tracing::error!("{name}: failed to open incident: {e}");
                                }
                            }
                            IncidentAction::Close => {
                                if let Err(e) = db::close_open_incidents(&pool, &name).await {
                                    tracing::error!("{name}: failed to close incidents: {e}");
                                }
                            }
                            IncidentAction::CloseAndOpen => {
                                if let Err(e) = db::close_and_open_incident(
                                    &pool,
                                    &name,
                                    &from_status.to_string(),
                                    &to_status.to_string(),
                                )
                                .await
                                {
                                    tracing::error!("{name}: failed to close+open incident: {e}");
                                }
                            }
                        }
                    }

                    // Latency drift detection
                    if let Some(ref trending) = trending_config
                        && snapshot.status == HealthStatus::Operational
                    {
                        let baseline_cutoff = (chrono::Utc::now()
                            - chrono::Duration::hours(trending.baseline_window_hours as i64))
                        .to_rfc3339();
                        let baseline_data = db::get_response_times(&pool, &name, &baseline_cutoff)
                            .await
                            .unwrap_or_default();
                        let operational_times: Vec<i64> = baseline_data
                            .iter()
                            .filter(|(_, ms)| *ms > 0)
                            .map(|(_, ms)| *ms)
                            .collect();
                        let baseline = LatencyStats::from_times(&operational_times);
                        let recent = db::get_recent_response_times(&pool, &name, 3)
                            .await
                            .unwrap_or_default();

                        if let Some(ref bl) = baseline {
                            if let Some(msg) =
                                drift::detect_latency_drift(&recent, bl, trending.spike_threshold)
                            {
                                if !in_drift {
                                    info!("{name}: {msg}");
                                    if let Some(ref alerter) = alerter {
                                        alerter.send_latency_drift_alert(&name, &label, &msg).await;
                                    }
                                    in_drift = true;
                                }
                            } else if in_drift {
                                info!("{name}: latency drift recovered");
                                if let Some(ref alerter) = alerter {
                                    alerter.send_latency_recovery(&name, &label).await;
                                }
                                in_drift = false;
                            }
                        }
                    }
                }
            }));
        }
    }

    // SSH banner checks, stored as health check entries with target name "{name}:ssh"
    for (name, target_config) in configured_targets(config) {
        if let Some(ssh_config) = target_config.ssh_banner {
            let interval_secs = default_interval;
            let pool = pool.clone();
            let check_name = format!("{name}:ssh");
            let label = target_config.label.clone();
            let alerter = alerter.cloned();
            let cancel = cancel.clone();

            info!("{check_name}: SSH banner check every {interval_secs}s");

            handles.push(tokio::spawn(async move {
                let mut ticks = CheckInterval::new(interval_secs, cancel);
                let seed = db::get_latest_health(&pool, &check_name)
                    .await
                    .ok()
                    .flatten()
                    .map(|s| s.status);
                let mut gate = TransitionGate::seeded(seed, confirmations);
                while ticks.next().await {
                    let snapshot = ssh_banner::check_ssh_banner(&check_name, &ssh_config).await;
                    info!(
                        "{}: {} ({}ms)",
                        check_name, snapshot.status, snapshot.response_time_ms
                    );
                    if let Err(e) = db::insert_health_check(&pool, &snapshot).await {
                        tracing::error!("{check_name}: failed to store SSH banner check: {e}");
                    }

                    if let Some((from_status, to_status)) = gate.observe(snapshot.status)
                        && let Some(ref alerter) = alerter
                    {
                        let from = from_status.to_string();
                        let to = to_status.to_string();
                        if to_status == HealthStatus::Operational {
                            alerter
                                .send_health_recovery(&check_name, &label, &from)
                                .await;
                        } else {
                            alerter
                                .send_health_alert(
                                    &check_name,
                                    &label,
                                    &from,
                                    &to,
                                    snapshot.error.as_deref(),
                                )
                                .await;
                        }
                    }
                }
            }));
        }
    }

    handles
}