Spaces:

stremio-db
/

a

Paused

FTA7700
Add cache restart-persistence and 7-day cleanup job
1ad3b12 3 months ago
24.4 kB
﻿use std::path::{Path, PathBuf};
use std::time::{Instant, SystemTime};

use sea_orm::*;
use tokio::fs;

pub mod persistence;

use crate::entity::{available_ratings, image_meta};
use crate::error::AppError;

pub struct CacheEntry {
    pub bytes: Vec<u8>,
    pub is_stale: bool,
}

#[derive(Clone)]
pub struct MemCacheEntry {
    pub bytes: bytes::Bytes,
    pub last_checked: Instant,
}

fn is_safe_path_component(s: &str) -> bool {
    !s.is_empty() && s != "." && s != ".." && !s.contains('/') && !s.contains('\\') && !s.contains('\0')
}

/// Reject id values that contain path traversal or null bytes.
/// Call this early — before any network calls — so malicious input
/// is caught as 400 rather than causing a downstream 500.
pub fn validate_id_value(id_value: &str) -> Result<(), AppError> {
    if !is_safe_path_component(id_value) {
        return Err(AppError::BadRequest("invalid id value".into()));
    }
    Ok(())
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum ImageType {
    Poster,
    Logo,
    Backdrop,
    Episode,
}

impl ImageType {
    pub fn subdir(self) -> &'static str {
        match self {
            ImageType::Poster => "posters",
            ImageType::Logo => "logos",
            ImageType::Backdrop => "backdrops",
            ImageType::Episode => "episodes",
        }
    }

    pub fn ext(self) -> &'static str {
        match self {
            ImageType::Poster | ImageType::Backdrop | ImageType::Episode => "jpg",
            ImageType::Logo => "png",
        }
    }

    pub fn db_value(self) -> &'static str {
        match self {
            ImageType::Poster => "p",
            ImageType::Logo => "l",
            ImageType::Backdrop => "b",
            ImageType::Episode => "e",
        }
    }

    /// Cache key prefix for image kind (poster has none).
    pub fn kind_prefix(self) -> &'static str {
        match self {
            ImageType::Poster => "",
            ImageType::Logo => "_l",
            ImageType::Backdrop => "_b",
            ImageType::Episode => "_e",
        }
    }

    /// Strip the file extension matching this image type from a string.
    pub fn strip_ext(self, s: &str) -> &str {
        match self {
            ImageType::Poster | ImageType::Backdrop | ImageType::Episode => s.strip_suffix(".jpg").unwrap_or(s),
            ImageType::Logo => s.strip_suffix(".png").unwrap_or(s),
        }
    }

    /// Human-readable label.
    pub fn label(self) -> &'static str {
        match self {
            ImageType::Poster => "poster",
            ImageType::Logo => "logo",
            ImageType::Backdrop => "backdrop",
            ImageType::Episode => "episode",
        }
    }

    pub fn content_type(self) -> &'static str {
        match self {
            ImageType::Poster | ImageType::Backdrop | ImageType::Episode => "image/jpeg",
            ImageType::Logo => "image/png",
        }
    }

}

/// Path for a rendered (composited) image: `{cache_dir}/{subdir}/{id_type}/{id_value}.{ext}`
pub fn typed_cache_path(
    cache_dir: &str,
    image_type: ImageType,
    id_type: &str,
    id_value: &str,
) -> Result<PathBuf, AppError> {
    if !is_safe_path_component(id_value) {
        return Err(AppError::BadRequest("invalid id value".into()));
    }
    if !is_safe_path_component(id_type) {
        return Err(AppError::BadRequest("invalid id type".into()));
    }
    let ext = image_type.ext();
    Ok(Path::new(cache_dir)
        .join(image_type.subdir())
        .join(id_type)
        .join(format!("{id_value}.{ext}")))
}

/// Path for a TMDB base poster: `{cache_dir}/base/posters/{tmdb_size}/{filename}`
pub fn base_poster_path(cache_dir: &str, poster_path: &str, tmdb_size: &str) -> Result<PathBuf, AppError> {
    // poster_path is like "/abc123.jpg" from TMDB
    let filename = poster_path.trim_start_matches('/');
    if !is_safe_path_component(filename) {
        return Err(AppError::BadRequest("invalid poster path".into()));
    }
    if !is_safe_path_component(tmdb_size) {
        return Err(AppError::BadRequest("invalid tmdb size".into()));
    }
    Ok(Path::new(cache_dir).join("base").join("posters").join(tmdb_size).join(filename))
}

/// Path for a fanart base image: `{cache_dir}/base/fanart/{fanart_id}.{ext}`
pub fn base_fanart_path(cache_dir: &str, fanart_id: &str, ext: &str) -> Result<PathBuf, AppError> {
    if !is_safe_path_component(fanart_id) {
        return Err(AppError::BadRequest("invalid fanart id".into()));
    }
    if !is_safe_path_component(ext) {
        return Err(AppError::BadRequest("invalid file extension".into()));
    }
    Ok(Path::new(cache_dir).join("base").join("fanart").join(format!("{fanart_id}.{ext}")))
}

/// Path for a preview image: `{cache_dir}/preview/{subdir}/{suffix}.{ext}`
pub fn preview_path(cache_dir: &str, image_type: ImageType, suffix: &str, ext: &str) -> Result<PathBuf, AppError> {
    if !is_safe_path_component(suffix) {
        return Err(AppError::BadRequest("invalid preview suffix".into()));
    }
    if !is_safe_path_component(ext) {
        return Err(AppError::BadRequest("invalid file extension".into()));
    }
    Ok(Path::new(cache_dir).join("preview").join(image_type.subdir()).join(format!("{suffix}.{ext}")))
}

/// Read a cached file. `stale_secs = 0` means never stale.
pub async fn read(path: &Path, stale_secs: u64) -> Option<CacheEntry> {
    let bytes = fs::read(path).await.ok()?;
    let metadata = fs::metadata(path).await.ok()?;
    let modified = metadata.modified().ok()?;
    let age = SystemTime::now()
        .duration_since(modified)
        .unwrap_or_default()
        .as_secs();

    Some(CacheEntry {
        bytes,
        is_stale: stale_secs > 0 && age > stale_secs,
    })
}

pub async fn write(path: &Path, bytes: &[u8]) -> Result<(), AppError> {
    if let Some(parent) = path.parent() {
        fs::create_dir_all(parent).await?;
    }
    fs::write(path, bytes).await?;
    Ok(())
}

pub async fn read_meta_db(db: &DatabaseConnection, cache_key: &str) -> Option<String> {
    image_meta::Entity::find_by_id(cache_key)
        .one(db)
        .await
        .ok()
        .flatten()
        .and_then(|m| m.release_date)
}

pub async fn read_meta_api_key_id(db: &DatabaseConnection, cache_key: &str) -> Option<i32> {
    image_meta::Entity::find_by_id(cache_key)
        .one(db)
        .await
        .ok()
        .flatten()
        .and_then(|m| m.api_key_id)
}

pub async fn upsert_meta_db(
    db: &DatabaseConnection,
    cache_key: &str,
    release_date: Option<&str>,
    image_type: ImageType,
    key_id: Option<i32>,
) -> Result<(), AppError> {
    let now = SystemTime::now()
        .duration_since(SystemTime::UNIX_EPOCH)
        .unwrap_or_default()
        .as_secs() as i64;

    let model = image_meta::ActiveModel {
        cache_key: Set(cache_key.to_string()),
        release_date: Set(release_date.map(|s| s.to_string())),
        image_type: Set(image_type.db_value().to_string()),
        created_at: Set(now),
        updated_at: Set(now),
        api_key_id: Set(key_id),
    };

    image_meta::Entity::insert(model)
        .on_conflict(
            sea_orm::sea_query::OnConflict::column(image_meta::Column::CacheKey)
                .update_columns([image_meta::Column::ReleaseDate, image_meta::Column::UpdatedAt, image_meta::Column::ApiKeyId])
                .to_owned(),
        )
        .exec(db)
        .await?;

    Ok(())
}

/// Read the stored available rating sources for a movie (e.g. `"ilrt"`).
///
/// Returns `None` if no entry exists yet, or if the stored data is stale
/// (using the same decay formula as poster/ratings caches: recent films
/// refresh frequently, films older than `max_age` never go stale).
pub async fn read_available_ratings(
    db: &DatabaseConnection,
    id_key: &str,
    min_stale: u64,
    max_age: u64,
) -> Option<String> {
    let model = available_ratings::Entity::find_by_id(id_key)
        .one(db)
        .await
        .ok()
        .flatten()?;

    let stale_secs = compute_stale_secs(model.release_date.as_deref(), min_stale, max_age);
    // stale_secs == 0 means "never stale" (old film), so always use the fast path
    if stale_secs > 0 {
        let now = SystemTime::now()
            .duration_since(SystemTime::UNIX_EPOCH)
            .unwrap_or_default()
            .as_secs();
        let age = now.saturating_sub(u64::try_from(model.updated_at).unwrap_or(0));
        if age > stale_secs {
            return None;
        }
    }

    Some(model.sources)
}

/// Store which rating sources have data for a movie so the hot path can
/// reconstruct the badges cache suffix without external API calls.
pub async fn upsert_available_ratings(
    db: &DatabaseConnection,
    id_key: &str,
    sources: &str,
    release_date: Option<&str>,
) -> Result<(), AppError> {
    let now = SystemTime::now()
        .duration_since(SystemTime::UNIX_EPOCH)
        .unwrap_or_default()
        .as_secs() as i64;

    let model = available_ratings::ActiveModel {
        id_key: Set(id_key.to_string()),
        sources: Set(sources.to_string()),
        updated_at: Set(now),
        release_date: Set(release_date.map(|s| s.to_string())),
    };

    available_ratings::Entity::insert(model)
        .on_conflict(
            sea_orm::sea_query::OnConflict::column(available_ratings::Column::IdKey)
                .update_columns([
                    available_ratings::Column::Sources,
                    available_ratings::Column::UpdatedAt,
                    available_ratings::Column::ReleaseDate,
                ])
                .to_owned(),
        )
        .exec(db)
        .await?;

    Ok(())
}

/// Parse "YYYY-MM-DD" to Unix epoch seconds. Returns `None` for invalid input.
fn date_str_to_epoch(s: &str) -> Option<u64> {
    let mut parts = s.split('-');
    let year: u64 = parts.next()?.parse().ok()?;
    let month: u64 = parts.next()?.parse().ok()?;
    let day: u64 = parts.next()?.parse().ok()?;
    if !(1..=12).contains(&month) || year < 1970 {
        return None;
    }
    let max_day = max_days_in_month(year, month);
    if !(1..=max_day).contains(&day) {
        return None;
    }

    Some(days_from_epoch(year, month, day) * 86400)
}

fn max_days_in_month(year: u64, month: u64) -> u64 {
    match month {
        1 | 3 | 5 | 7 | 8 | 10 | 12 => 31,
        4 | 6 | 9 | 11 => 30,
        2 => if is_leap(year) { 29 } else { 28 },
        _ => 0,
    }
}

/// Days from 1970-01-01 to the given date, using a closed-form leap year count.
fn days_from_epoch(year: u64, month: u64, day: u64) -> u64 {
    // Leap years in [1, y] = y/4 - y/100 + y/400
    let leaps_before = |y: u64| y / 4 - y / 100 + y / 400;
    let prev = year - 1;
    let days_to_year = 365 * (year - 1970) + leaps_before(prev) - leaps_before(1969);

    let days_in_month = [0, 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31];
    let mut month_days: u64 = 0;
    for m in 1..month {
        month_days += days_in_month[m as usize] as u64;
    }
    if month > 2 && is_leap(year) {
        month_days += 1;
    }

    days_to_year + month_days + day - 1
}

fn is_leap(y: u64) -> bool {
    (y % 4 == 0 && y % 100 != 0) || y % 400 == 0
}

/// Compute dynamic stale_secs based on release date.
/// Returns 0 (never stale) for films older than `max_age`.
pub fn compute_stale_secs(
    release_date_str: Option<&str>,
    min_stale: u64,
    max_age: u64,
) -> u64 {
    let now = SystemTime::now()
        .duration_since(SystemTime::UNIX_EPOCH)
        .unwrap_or_default()
        .as_secs();

    let epoch = match release_date_str.and_then(date_str_to_epoch) {
        Some(e) => e,
        None => return min_stale,
    };

    if epoch > now {
        // Unreleased / future film
        return min_stale;
    }

    let film_age = now - epoch;
    if film_age >= max_age {
        return 0; // never stale
    }

    // Linear interpolation: min_stale at age=0, approaches max_age at age=max_age
    min_stale + film_age * (max_age - min_stale) / max_age
}

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

    #[test]
    fn compute_stale_no_release_date() {
        let result = compute_stale_secs(None, 86400, 31_536_000);
        assert_eq!(result, 86400);
    }

    #[test]
    fn compute_stale_invalid_date() {
        let result = compute_stale_secs(Some("not-a-date"), 86400, 31_536_000);
        assert_eq!(result, 86400);
    }

    #[test]
    fn compute_stale_future_film() {
        let result = compute_stale_secs(Some("2099-01-01"), 86400, 31_536_000);
        assert_eq!(result, 86400);
    }

    #[test]
    fn compute_stale_old_film() {
        // Film from 2000 — age far exceeds max_age of 1 year
        let result = compute_stale_secs(Some("2000-01-01"), 86400, 31_536_000);
        assert_eq!(result, 0);
    }

    #[test]
    fn date_str_to_epoch_known_value() {
        // 1970-01-02 should be exactly 86400 seconds
        assert_eq!(date_str_to_epoch("1970-01-02"), Some(86400));
    }

    #[test]
    fn date_str_to_epoch_epoch_start() {
        assert_eq!(date_str_to_epoch("1970-01-01"), Some(0));
    }

    #[test]
    fn date_str_to_epoch_invalid_month() {
        assert_eq!(date_str_to_epoch("2020-13-01"), None);
    }

    #[test]
    fn date_str_to_epoch_invalid_day() {
        assert_eq!(date_str_to_epoch("2020-01-32"), None);
    }

    #[test]
    fn date_str_to_epoch_pre_epoch() {
        assert_eq!(date_str_to_epoch("1969-01-01"), None);
    }

    #[test]
    fn date_str_to_epoch_feb_30_rejected() {
        assert_eq!(date_str_to_epoch("2020-02-30"), None);
    }

    #[test]
    fn date_str_to_epoch_feb_29_leap_accepted() {
        assert!(date_str_to_epoch("2020-02-29").is_some());
    }

    #[test]
    fn date_str_to_epoch_feb_29_non_leap_rejected() {
        assert_eq!(date_str_to_epoch("2023-02-29"), None);
    }

    #[test]
    fn date_str_to_epoch_apr_31_rejected() {
        assert_eq!(date_str_to_epoch("2020-04-31"), None);
    }

    #[test]
    fn date_str_to_epoch_garbage() {
        assert_eq!(date_str_to_epoch("garbage"), None);
    }

    #[test]
    fn typed_cache_path_poster() {
        let p = typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", "tt1234567").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/posters/imdb/tt1234567.jpg"));
    }

    #[test]
    fn typed_cache_path_logo() {
        let p = typed_cache_path("/tmp/cache", ImageType::Logo, "imdb", "tt1234567").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/logos/imdb/tt1234567.png"));
    }

    #[test]
    fn typed_cache_path_backdrop() {
        let p = typed_cache_path("/tmp/cache", ImageType::Backdrop, "imdb", "tt1234567").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/backdrops/imdb/tt1234567.jpg"));
    }

    #[test]
    fn typed_cache_path_rejects_traversal() {
        assert!(typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", "../../etc/passwd").is_err());
        assert!(typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", "..").is_err());
        assert!(typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", ".").is_err());
        assert!(typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", "").is_err());
        assert!(typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", "foo/bar").is_err());
        assert!(typed_cache_path("/tmp/cache", ImageType::Poster, "imdb", "foo\\bar").is_err());
    }

    #[test]
    fn base_poster_path_strips_leading_slash() {
        let p = base_poster_path("/tmp/cache", "/abc123.jpg", "w500").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/base/posters/w500/abc123.jpg"));
    }

    #[test]
    fn base_poster_path_no_leading_slash() {
        let p = base_poster_path("/tmp/cache", "abc123.jpg", "w500").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/base/posters/w500/abc123.jpg"));
    }

    #[test]
    fn base_poster_path_original_size() {
        let p = base_poster_path("/tmp/cache", "/abc123.jpg", "original").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/base/posters/original/abc123.jpg"));
    }

    #[test]
    fn base_poster_path_rejects_traversal() {
        assert!(base_poster_path("/tmp/cache", "/../etc/passwd", "w500").is_err());
        assert!(base_poster_path("/tmp/cache", "..", "w500").is_err());
        assert!(base_poster_path("/tmp/cache", "", "w500").is_err());
    }

    #[test]
    fn base_poster_path_rejects_invalid_tmdb_size() {
        assert!(base_poster_path("/tmp/cache", "abc123.jpg", "..").is_err());
        assert!(base_poster_path("/tmp/cache", "abc123.jpg", "").is_err());
        assert!(base_poster_path("/tmp/cache", "abc123.jpg", "../etc").is_err());
    }

    #[test]
    fn base_fanart_path_valid() {
        let p = base_fanart_path("/tmp/cache", "12345", "png").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/base/fanart/12345.png"));
    }

    #[test]
    fn base_fanart_path_rejects_traversal() {
        assert!(base_fanart_path("/tmp/cache", "..", "png").is_err());
        assert!(base_fanart_path("/tmp/cache", "12345", "..").is_err());
        assert!(base_fanart_path("/tmp/cache", "", "png").is_err());
        assert!(base_fanart_path("/tmp/cache", "foo/bar", "png").is_err());
    }

    #[test]
    fn preview_path_valid() {
        let p = preview_path("/tmp/cache", ImageType::Poster, "r_imdb", "jpg").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/preview/posters/r_imdb.jpg"));
    }

    #[test]
    fn preview_path_logo() {
        let p = preview_path("/tmp/cache", ImageType::Logo, "r_imdb", "png").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/preview/logos/r_imdb.png"));
    }

    #[test]
    fn preview_path_backdrop() {
        let p = preview_path("/tmp/cache", ImageType::Backdrop, "r_imdb", "jpg").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/preview/backdrops/r_imdb.jpg"));
    }

    #[test]
    fn preview_path_rejects_traversal() {
        assert!(preview_path("/tmp/cache", ImageType::Poster, "..", "jpg").is_err());
        assert!(preview_path("/tmp/cache", ImageType::Poster, "", "jpg").is_err());
        assert!(preview_path("/tmp/cache", ImageType::Poster, "foo", "..").is_err());
    }

    #[test]
    fn image_type_subdir() {
        assert_eq!(ImageType::Poster.subdir(), "posters");
        assert_eq!(ImageType::Logo.subdir(), "logos");
        assert_eq!(ImageType::Backdrop.subdir(), "backdrops");
        assert_eq!(ImageType::Episode.subdir(), "episodes");
    }

    #[test]
    fn image_type_ext() {
        assert_eq!(ImageType::Poster.ext(), "jpg");
        assert_eq!(ImageType::Logo.ext(), "png");
        assert_eq!(ImageType::Backdrop.ext(), "jpg");
        assert_eq!(ImageType::Episode.ext(), "jpg");
    }

    #[test]
    fn image_type_episode_properties() {
        assert_eq!(ImageType::Episode.db_value(), "e");
        assert_eq!(ImageType::Episode.kind_prefix(), "_e");
        assert_eq!(ImageType::Episode.label(), "episode");
        assert_eq!(ImageType::Episode.content_type(), "image/jpeg");
        assert_eq!(ImageType::Episode.strip_ext("still.jpg"), "still");
        assert_eq!(ImageType::Episode.strip_ext("still.png"), "still.png");
    }

    #[test]
    fn typed_cache_path_episode() {
        let p = typed_cache_path("/tmp/cache", ImageType::Episode, "tmdb", "episode-1396-S1E1").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/episodes/tmdb/episode-1396-S1E1.jpg"));
    }

    #[test]
    fn preview_path_episode() {
        let p = preview_path("/tmp/cache", ImageType::Episode, "r_imdb", "jpg").unwrap();
        assert_eq!(p, PathBuf::from("/tmp/cache/preview/episodes/r_imdb.jpg"));
    }

    #[test]
    fn is_leap_year_cases() {
        assert!(is_leap(2000)); // divisible by 400
        assert!(is_leap(2024)); // divisible by 4, not by 100
        assert!(!is_leap(1900)); // divisible by 100, not by 400
        assert!(!is_leap(2023)); // not divisible by 4
    }
}

/// Deletes every `image_meta` row whose `cache_key` starts with any of the
/// given prefixes (e.g. "imdb/tt1234567", "tmdb/movie-5678"). Since
/// `EXTERNAL_CACHE_ONLY=true` means the Postgres row *is* the cache, deleting
/// it is sufficient to force a fresh render on next request. Returns the
/// number of rows deleted.
pub async fn purge_by_id_prefixes(
    db: &sea_orm::DatabaseConnection,
    prefixes: &[String],
) -> Result<u64, AppError> {
    use crate::entity::image_meta;
    use sea_orm::{ColumnTrait, Condition, QueryFilter, EntityTrait};
    if prefixes.is_empty() {
        return Ok(0);
    }
    let mut cond = Condition::any();
    for p in prefixes {
        cond = cond.add(image_meta::Column::CacheKey.starts_with(p.as_str()));
    }
    let res = image_meta::Entity::delete_many()
        .filter(cond)
        .exec(db)
        .await
        .map_err(|e| AppError::DbError(e.to_string()))?;
    Ok(res.rows_affected)
}
/// Deletes matching rows from `available_ratings` by exact `id_key` match
/// (the table stores bare ids like "imdb/tt1234567" with no settings-hash
/// suffix, unlike `image_meta.cache_key`). Needed alongside
/// `purge_by_id_prefixes`: without this, a purged title can still take the
/// "fast path" on next render and reuse stale rating values pulled straight
/// from this table, even though its image_meta row is gone.
pub async fn purge_available_ratings(
    db: &sea_orm::DatabaseConnection,
    ids: &[String],
) -> Result<u64, AppError> {
    use sea_orm::{ConnectionTrait, Statement};
    if ids.is_empty() {
        return Ok(0);
    }
    let placeholders: Vec<String> = (1..=ids.len()).map(|i| format!("${i}")).collect();
    let sql = format!(
        "DELETE FROM available_ratings WHERE id_key IN ({})",
        placeholders.join(", ")
    );
    let values: Vec<sea_orm::Value> = ids.iter().map(|s| s.clone().into()).collect();
    let stmt = Statement::from_sql_and_values(db.get_database_backend(), &sql, values);
    let res = db
        .execute(stmt)
        .await
        .map_err(|e| AppError::DbError(e.to_string()))?;
    Ok(res.rows_affected())
}

/// Returns the subset of `keys` that currently exist as `cache_key` values in
/// `image_meta`. Used by the Library staleness checker: given the set of
/// cache_keys a Library item *would* have if rendered under current settings,
/// this tells us which items already have a matching render (fresh) and which
/// don't (stale — only old-settings renders exist, or none at all).
pub async fn find_existing_cache_keys(
    db: &sea_orm::DatabaseConnection,
    keys: &[String],
) -> Result<std::collections::HashSet<String>, AppError> {
    use crate::entity::image_meta;
    use sea_orm::{ColumnTrait, EntityTrait, QueryFilter, QuerySelect};
    if keys.is_empty() {
        return Ok(std::collections::HashSet::new());
    }
    let rows = image_meta::Entity::find()
        .filter(image_meta::Column::CacheKey.is_in(keys.iter().cloned()))
        .select_only()
        .column(image_meta::Column::CacheKey)
        .into_tuple::<String>()
        .all(db)
        .await
        .map_err(|e| AppError::DbError(e.to_string()))?;
    Ok(rows.into_iter().collect())
}

/// For each given ID prefix (e.g. "imdb/tt1234567"), returns the most
/// recent `updated_at` across all `image_meta` rows whose `cache_key`
/// starts with that prefix (i.e. across every settings-suffix variant ever
/// rendered for that ID). Used by the staleness checker to determine when
/// an item was last rendered, regardless of which settings suffix it landed
/// under at the time.
pub async fn max_updated_at_by_prefix(
    db: &sea_orm::DatabaseConnection,
    prefixes: &[String],
) -> Result<std::collections::HashMap<String, i64>, AppError> {
    use crate::entity::image_meta;
    use sea_orm::{ColumnTrait, EntityTrait, QueryFilter, QuerySelect};
    let mut result = std::collections::HashMap::new();
    if prefixes.is_empty() {
        return Ok(result);
    }
    for prefix in prefixes {
        let rows = image_meta::Entity::find()
            .filter(image_meta::Column::CacheKey.starts_with(prefix.as_str()))
            .select_only()
            .column(image_meta::Column::UpdatedAt)
            .into_tuple::<i64>()
            .all(db)
            .await
            .map_err(|e| AppError::DbError(e.to_string()))?;
        if let Some(max) = rows.into_iter().max() {
            result.insert(prefix.clone(), max);
        }
    }
    Ok(result)
}