Files

505 lines
16 KiB
Go
Raw Permalink Normal View History

package store
import (
"context"
"encoding/json"
"fmt"
"strings"
"time"
"github.com/jackc/pgx/v5"
)
// LibraryItem is one imported Emby item. Payload is Emby's JSON verbatim; the flat
// columns exist only so Postgres can filter and rank without opening the JSON.
type LibraryItem struct {
ID string
Type string
Name string
SeriesID string
SeriesName string
ProductionYear *int
CommunityRating *float64
Genres []string
Studios []string
DateCreated *time.Time
SearchText string
Payload json.RawMessage
}
// LibraryStats is what the admin page shows about the imported library.
type LibraryStats struct {
Total int64 `json:"total"`
ByType map[string]int64 `json:"byType"`
LastSynced *time.Time `json:"lastSynced"`
}
2026-08-02 22:10:19 +12:00
// UpsertLibraryItems writes a batch, refreshing synced_at on every row it touches, and
// returns how many recommendation-relevant payloads were inserted or actually changed.
//
// synced_at doubles as the mark-and-sweep marker: a full import stamps everything it
// sees, then deletes whatever kept an older stamp.
func (s *Store) UpsertLibraryItems(ctx context.Context, items []LibraryItem, syncedAt time.Time) (int64, error) {
if len(items) == 0 {
return 0, nil
}
batch := &pgx.Batch{}
for _, item := range items {
batch.Queue(`
2026-08-02 22:10:19 +12:00
WITH previous AS MATERIALIZED (
SELECT type, name, series_id, series_name, production_year,
community_rating, genres, studios, date_created,
payload->'RunTimeTicks' AS runtime_ticks,
payload->'MediaStreams' AS media_streams,
payload->'Container' AS container
FROM library_items WHERE id = $1
), upserted AS (
INSERT INTO library_items (
id, type, name, series_id, series_name, production_year, community_rating,
genres, studios, date_created, search_text, payload, synced_at
) VALUES ($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12::jsonb,$13)
ON CONFLICT (id) DO UPDATE SET
type = EXCLUDED.type,
name = EXCLUDED.name,
series_id = EXCLUDED.series_id,
series_name = EXCLUDED.series_name,
production_year = EXCLUDED.production_year,
community_rating = EXCLUDED.community_rating,
genres = EXCLUDED.genres,
studios = EXCLUDED.studios,
date_created = EXCLUDED.date_created,
search_text = EXCLUDED.search_text,
payload = EXCLUDED.payload,
synced_at = EXCLUDED.synced_at
RETURNING 1
)
SELECT NOT EXISTS (SELECT 1 FROM previous)
OR EXISTS (
SELECT 1 FROM previous
WHERE ROW(
type, name, series_id, series_name, production_year,
community_rating, genres, studios, date_created,
runtime_ticks, media_streams, container
) IS DISTINCT FROM ROW(
$2::text, $3::text, $4::text, $5::text, $6::int,
$7::real, $8::text[], $9::text[], $10::timestamptz,
$12::jsonb->'RunTimeTicks', $12::jsonb->'MediaStreams',
$12::jsonb->'Container'
)
)
FROM upserted`,
item.ID, item.Type, item.Name, item.SeriesID, item.SeriesName,
item.ProductionYear, item.CommunityRating, item.Genres, item.Studios,
item.DateCreated, item.SearchText, string(item.Payload), syncedAt)
}
results := s.pool.SendBatch(ctx, batch)
defer results.Close()
2026-08-02 22:10:19 +12:00
var changed int64
for range items {
2026-08-02 22:10:19 +12:00
var recommendationChanged bool
if err := results.QueryRow().Scan(&recommendationChanged); err != nil {
return changed, fmt.Errorf("store: upsert library items: %w", err)
}
if recommendationChanged {
changed++
}
}
2026-08-02 22:10:19 +12:00
return changed, nil
}
// DeleteLibraryItemsBefore removes anything a full import did not touch — items deleted
// from Emby since the last run.
func (s *Store) DeleteLibraryItemsBefore(ctx context.Context, cutoff time.Time) (int64, error) {
tag, err := s.pool.Exec(ctx, `DELETE FROM library_items WHERE synced_at < $1`, cutoff)
if err != nil {
return 0, fmt.Errorf("store: prune library: %w", err)
}
return tag.RowsAffected(), nil
}
2026-08-18 14:59:29 +12:00
// NamedItem is the least a caller can be told about a catalogue row and still identify
// it: what it is called and, where the library knows, when it came out.
type NamedItem struct {
ID string
Name string
Year int
}
// LibraryItemsByName finds catalogue rows by title, case-insensitively.
//
// The comparison that decides the answer is not this one: the caller normalises both
// sides (punctuation and spacing are where an *arr and Emby actually differ) and picks by
// year. This is the narrowing query — a handful of rows out of twenty thousand — so that
// the matching rule can stay a pure function with one definition.
func (s *Store) LibraryItemsByName(ctx context.Context, itemType, name string) ([]NamedItem, error) {
trimmed := strings.TrimSpace(name)
if trimmed == "" {
return nil, nil
}
rows, err := s.pool.Query(ctx, `
SELECT id, name, COALESCE(production_year, 0)
FROM library_items
WHERE type = $1 AND lower(name) = lower($2)
LIMIT 50`, itemType, trimmed)
if err != nil {
return nil, fmt.Errorf("store: library items by name: %w", err)
}
defer rows.Close()
out := []NamedItem{}
for rows.Next() {
var item NamedItem
if err := rows.Scan(&item.ID, &item.Name, &item.Year); err != nil {
return nil, err
}
out = append(out, item)
}
return out, rows.Err()
}
// DeleteLibraryItem removes one item and anything derived from it.
//
// The credits marker goes with it, and that is the point of doing this in one place: the
// marker table is keyed on the item id and nothing else prunes it, so a title deleted from
// the library would otherwise leave a Skip Credits position behind for a file that no
// longer exists — and if that id were ever reused, in front of the wrong programme.
//
// Deleting a series takes its episodes with it, because Emby's own hierarchy is the only
// thing that made those rows meaningful.
func (s *Store) DeleteLibraryItem(ctx context.Context, itemID string) (int64, error) {
if strings.TrimSpace(itemID) == "" {
return 0, nil
}
tag, err := s.pool.Exec(ctx,
`DELETE FROM library_items WHERE id = $1 OR series_id = $1`, itemID)
if err != nil {
return 0, fmt.Errorf("store: delete library item: %w", err)
}
if _, err := s.pool.Exec(ctx,
`DELETE FROM credits_markers WHERE item_id = $1 OR series_id = $1`, itemID); err != nil {
return 0, fmt.Errorf("store: delete credits markers: %w", err)
}
return tag.RowsAffected(), nil
}
// SearchLibrary answers from the imported library rather than Emby.
//
// Full-text match first, with a trailing ILIKE so partial words ("sever") still hit
// before someone finishes typing on a remote.
func (s *Store) SearchLibrary(ctx context.Context, term string, limit int) ([]json.RawMessage, error) {
trimmed := strings.TrimSpace(term)
if trimmed == "" {
return nil, nil
}
rows, err := s.pool.Query(ctx, `
SELECT payload
FROM library_items
WHERE search_tsv @@ plainto_tsquery('simple', $1)
OR search_text ILIKE '%' || $1 || '%'
ORDER BY
ts_rank(search_tsv, plainto_tsquery('simple', $1)) DESC,
(lower(name) = lower($1)) DESC,
community_rating DESC NULLS LAST,
name ASC
LIMIT $2`, trimmed, limit)
if err != nil {
return nil, fmt.Errorf("store: search library: %w", err)
}
return collectPayloads(rows)
}
2026-08-12 08:25:15 +12:00
// LibraryContainsProviderIDs reports which external catalogue ids are already represented in
// Emby. ProviderIds stays inside the imported payload because it is not a ranking field;
// request autocomplete is the one place that needs to compare it with Radarr/Sonarr.
func (s *Store) LibraryContainsProviderIDs(
ctx context.Context, provider string, ids []int,
) (map[int]bool, error) {
found := map[int]bool{}
if len(ids) == 0 {
return found, nil
}
values := make([]string, 0, len(ids))
for _, id := range ids {
if id > 0 {
values = append(values, fmt.Sprint(id))
}
}
rows, err := s.pool.Query(ctx, `
SELECT DISTINCT (payload->'ProviderIds'->>$1)::int
FROM library_items
WHERE payload->'ProviderIds'->>$1 = ANY($2::text[])`, provider, values)
if err != nil {
return nil, fmt.Errorf("store: library provider ids: %w", err)
}
defer rows.Close()
for rows.Next() {
var id int
if err := rows.Scan(&id); err != nil {
return nil, err
}
found[id] = true
}
return found, rows.Err()
}
2026-08-12 14:13:19 +12:00
// LibraryProviderItemIDs is LibraryContainsProviderIDs with the answer the request page
// needs: not only whether Emby has the title, but which item it is.
//
// That id is what lets a request that has finally downloaded stop being a status card and
// become something a viewer can press. Ordering by item_id keeps the answer stable when a
// household holds the same title twice — a duplicate import, or a remake sharing an id in
// bad metadata — so a card does not point at a different copy between two reads.
func (s *Store) LibraryProviderItemIDs(
ctx context.Context, provider string, ids []int,
) (map[int]string, error) {
found := map[int]string{}
if len(ids) == 0 {
return found, nil
}
values := make([]string, 0, len(ids))
for _, id := range ids {
if id > 0 {
values = append(values, fmt.Sprint(id))
}
}
if len(values) == 0 {
return found, nil
}
rows, err := s.pool.Query(ctx, `
SELECT DISTINCT ON ((payload->'ProviderIds'->>$1)::int)
(payload->'ProviderIds'->>$1)::int, id
FROM library_items
WHERE payload->'ProviderIds'->>$1 = ANY($2::text[])
ORDER BY (payload->'ProviderIds'->>$1)::int, id`, provider, values)
if err != nil {
return nil, fmt.Errorf("store: library provider item ids: %w", err)
}
defer rows.Close()
for rows.Next() {
var (
id int
itemID string
)
if err := rows.Scan(&id, &itemID); err != nil {
return nil, err
}
found[id] = itemID
}
return found, rows.Err()
}
// LibraryCandidates returns unwatched-agnostic candidates in the given genres, for the
// recommendation engine. User state is applied by the caller, which is the only place
// that knows it.
func (s *Store) LibraryCandidates(ctx context.Context, genres []string, limit int) ([]json.RawMessage, error) {
if len(genres) == 0 {
return nil, nil
}
rows, err := s.pool.Query(ctx, `
SELECT payload
FROM library_items
WHERE type IN ('Movie', 'Series')
AND genres && $1
ORDER BY community_rating DESC NULLS LAST, date_created DESC NULLS LAST
LIMIT $2`, genres, limit)
if err != nil {
return nil, fmt.Errorf("store: library candidates: %w", err)
}
return collectPayloads(rows)
}
2026-07-29 15:26:27 +12:00
// AllRecommendationCandidates returns the complete Movie/Series catalogue for an
// offline For You rebuild. The resulting per-user pool is deliberately over-provisioned
// so a short runtime filter still has enough ranked titles to fill the TV row.
func (s *Store) AllRecommendationCandidates(ctx context.Context) ([]json.RawMessage, error) {
rows, err := s.pool.Query(ctx, `
SELECT payload
FROM library_items
WHERE type IN ('Movie', 'Series')
ORDER BY community_rating DESC NULLS LAST, date_created DESC NULLS LAST`)
if err != nil {
return nil, fmt.Errorf("store: all recommendation candidates: %w", err)
}
return collectPayloads(rows)
}
2026-08-02 22:10:19 +12:00
func (s *Store) LibraryItemsByID(
ctx context.Context,
ids []string,
) ([]json.RawMessage, error) {
if len(ids) == 0 {
return []json.RawMessage{}, nil
}
rows, err := s.pool.Query(ctx, `
SELECT payload
FROM library_items
WHERE id = ANY($1)`, ids)
if err != nil {
return nil, fmt.Errorf("store: library items by id: %w", err)
}
return collectPayloads(rows)
}
2026-07-27 21:06:51 +12:00
// CuratedCandidates filters the imported catalogue for a server-authored shelf. Arrays
// are matched case-insensitively because Emby studio capitalisation is not consistent.
func (s *Store) CuratedCandidates(
ctx context.Context,
itemTypes, genres, studios []string,
limit int,
) ([]json.RawMessage, error) {
if len(itemTypes) == 0 || (len(genres) == 0 && len(studios) == 0) {
return nil, nil
}
rows, err := s.pool.Query(ctx, `
SELECT payload
FROM library_items
WHERE type = ANY($1)
AND (
cardinality($2::text[]) = 0 OR
EXISTS (
SELECT 1 FROM unnest(genres) AS genre
WHERE lower(genre) = ANY($2)
)
)
AND (
cardinality($3::text[]) = 0 OR
EXISTS (
SELECT 1 FROM unnest(studios) AS studio
WHERE lower(studio) = ANY($3)
)
)
ORDER BY community_rating DESC NULLS LAST, date_created DESC NULLS LAST
LIMIT $4`,
itemTypes, lowerStrings(genres), lowerStrings(studios), limit)
if err != nil {
return nil, fmt.Errorf("store: curated candidates: %w", err)
}
return collectPayloads(rows)
}
2026-08-02 22:10:19 +12:00
// LibraryGenres returns every genre with enough catalogue depth to make a shelf feel
// intentional. The recommendation engine still applies per-user seen filtering and may
// drop a shelf afterwards when too few unseen titles remain.
func (s *Store) LibraryGenres(
ctx context.Context,
itemTypes []string,
minItems int,
) ([]string, error) {
if len(itemTypes) == 0 {
return nil, nil
}
if minItems < 1 {
minItems = 1
}
rows, err := s.pool.Query(ctx, `
SELECT genre, count(DISTINCT item.id) AS item_count
FROM library_items AS item
CROSS JOIN LATERAL unnest(item.genres) AS genre
WHERE item.type = ANY($1)
AND btrim(genre) <> ''
GROUP BY genre
HAVING count(DISTINCT item.id) >= $2
ORDER BY item_count DESC, lower(genre) ASC`,
itemTypes, minItems)
if err != nil {
return nil, fmt.Errorf("store: library genres: %w", err)
}
defer rows.Close()
out := []string{}
for rows.Next() {
var genre string
var count int
if err := rows.Scan(&genre, &count); err != nil {
return nil, err
}
out = append(out, genre)
}
return out, rows.Err()
}
2026-08-06 22:33:56 +12:00
// SeriesRef is the minimum needed to link an outside catalogue's show — Sonarr's, in
// practice — to the Emby series the library holds, so a card built from that catalogue
// can open the show's own page. Year is 0 when Emby does not know it.
type SeriesRef struct {
2026-08-11 12:08:51 +12:00
ID string
Name string
Year int
LogoTag string
2026-08-06 22:33:56 +12:00
}
// SeriesRefs lists every imported series. The catalogue is a household's, not a
// provider's: a few hundred rows of three short columns, which is why this reads them
// all rather than querying per title.
func (s *Store) SeriesRefs(ctx context.Context) ([]SeriesRef, error) {
rows, err := s.pool.Query(ctx, `
2026-08-11 12:08:51 +12:00
SELECT id, name, COALESCE(production_year, 0),
COALESCE(payload->'ImageTags'->>'Logo', '')
2026-08-06 22:33:56 +12:00
FROM library_items
WHERE type = 'Series'`)
if err != nil {
return nil, fmt.Errorf("store: series refs: %w", err)
}
defer rows.Close()
out := []SeriesRef{}
for rows.Next() {
var ref SeriesRef
2026-08-11 12:08:51 +12:00
if err := rows.Scan(&ref.ID, &ref.Name, &ref.Year, &ref.LogoTag); err != nil {
2026-08-06 22:33:56 +12:00
return nil, err
}
out = append(out, ref)
}
return out, rows.Err()
}
2026-07-27 21:06:51 +12:00
func lowerStrings(values []string) []string {
out := make([]string, 0, len(values))
for _, value := range values {
out = append(out, strings.ToLower(strings.TrimSpace(value)))
}
return out
}
func (s *Store) LibraryStats(ctx context.Context) (LibraryStats, error) {
stats := LibraryStats{ByType: map[string]int64{}}
rows, err := s.pool.Query(ctx, `SELECT type, count(*) FROM library_items GROUP BY type`)
if err != nil {
return stats, fmt.Errorf("store: library stats: %w", err)
}
defer rows.Close()
for rows.Next() {
var itemType string
var count int64
if err := rows.Scan(&itemType, &count); err != nil {
return stats, err
}
stats.ByType[itemType] = count
stats.Total += count
}
if err := rows.Err(); err != nil {
return stats, err
}
var lastSynced *time.Time
if err := s.pool.QueryRow(ctx, `SELECT max(synced_at) FROM library_items`).Scan(&lastSynced); err != nil {
return stats, err
}
stats.LastSynced = lastSynced
return stats, nil
}
func collectPayloads(rows pgx.Rows) ([]json.RawMessage, error) {
defer rows.Close()
out := []json.RawMessage{}
for rows.Next() {
var payload []byte
if err := rows.Scan(&payload); err != nil {
return nil, err
}
out = append(out, json.RawMessage(payload))
}
return out, rows.Err()
}