222 lines
11 KiB
SQL
222 lines
11 KiB
SQL
-- Gateway sessions: one row per signed-in TV.
|
|
--
|
|
-- token_hash is SHA-256 of the bearer token handed to the device, so a database dump
|
|
-- does not hand over working gateway tokens. emby_token IS the live upstream token and
|
|
-- is stored as-is: treat this volume as a secret store.
|
|
CREATE TABLE IF NOT EXISTS sessions (
|
|
token_hash BYTEA PRIMARY KEY,
|
|
emby_user_id TEXT NOT NULL,
|
|
emby_token TEXT NOT NULL,
|
|
username TEXT NOT NULL,
|
|
server_id TEXT NOT NULL DEFAULT '',
|
|
device_id TEXT NOT NULL DEFAULT '',
|
|
device_name TEXT NOT NULL DEFAULT 'Memby TV',
|
|
client_version TEXT NOT NULL DEFAULT '',
|
|
client_protocol TEXT NOT NULL DEFAULT '',
|
|
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
last_seen_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
|
);
|
|
|
|
ALTER TABLE sessions ADD COLUMN IF NOT EXISTS device_name TEXT NOT NULL DEFAULT 'Memby TV';
|
|
ALTER TABLE sessions ADD COLUMN IF NOT EXISTS client_version TEXT NOT NULL DEFAULT '';
|
|
ALTER TABLE sessions ADD COLUMN IF NOT EXISTS client_protocol TEXT NOT NULL DEFAULT '';
|
|
|
|
-- Older builds could create more than one token for the same physical TV. Keep the most
|
|
-- recently used row before adding the identity constraint.
|
|
DELETE FROM sessions older
|
|
USING sessions newer
|
|
WHERE older.emby_user_id = newer.emby_user_id
|
|
AND older.device_id = newer.device_id
|
|
AND (
|
|
older.last_seen_at < newer.last_seen_at
|
|
OR (older.last_seen_at = newer.last_seen_at AND older.token_hash < newer.token_hash)
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS sessions_emby_user_idx ON sessions (emby_user_id);
|
|
CREATE INDEX IF NOT EXISTS sessions_last_seen_idx ON sessions (last_seen_at);
|
|
CREATE UNIQUE INDEX IF NOT EXISTS sessions_user_device_idx
|
|
ON sessions (emby_user_id, device_id);
|
|
|
|
-- The imported library.
|
|
--
|
|
-- payload is Emby's item JSON verbatim, so rows served from here are byte-identical to
|
|
-- rows served live. Deliberately holds NO per-user state: everything is imported with
|
|
-- EnableUserData=false, because one household shares this table and watched/favourite
|
|
-- flags are not shareable. Anything user-specific still comes from Emby live.
|
|
CREATE TABLE IF NOT EXISTS library_items (
|
|
id TEXT PRIMARY KEY,
|
|
type TEXT NOT NULL DEFAULT '',
|
|
name TEXT NOT NULL DEFAULT '',
|
|
series_id TEXT NOT NULL DEFAULT '',
|
|
series_name TEXT NOT NULL DEFAULT '',
|
|
production_year INT,
|
|
community_rating REAL,
|
|
genres TEXT[] NOT NULL DEFAULT '{}',
|
|
studios TEXT[] NOT NULL DEFAULT '{}',
|
|
date_created TIMESTAMPTZ,
|
|
search_text TEXT NOT NULL DEFAULT '',
|
|
payload JSONB NOT NULL,
|
|
synced_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
-- 'simple' rather than 'english': film titles are proper nouns, and stemming
|
|
-- "Arrival" into "arriv" helps nobody.
|
|
search_tsv tsvector GENERATED ALWAYS AS (to_tsvector('simple', search_text)) STORED
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS library_items_search_idx ON library_items USING GIN (search_tsv);
|
|
CREATE INDEX IF NOT EXISTS library_items_genres_idx ON library_items USING GIN (genres);
|
|
CREATE INDEX IF NOT EXISTS library_items_type_created_idx ON library_items (type, date_created DESC);
|
|
CREATE INDEX IF NOT EXISTS library_items_synced_idx ON library_items (synced_at);
|
|
|
|
-- One row per import, so the admin page can show what happened and when.
|
|
CREATE TABLE IF NOT EXISTS sync_runs (
|
|
id BIGSERIAL PRIMARY KEY,
|
|
kind TEXT NOT NULL, -- full | incremental
|
|
trigger TEXT NOT NULL DEFAULT 'schedule', -- schedule | manual | startup
|
|
status TEXT NOT NULL, -- running | success | failed
|
|
started_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
finished_at TIMESTAMPTZ,
|
|
items_seen INT NOT NULL DEFAULT 0,
|
|
items_upserted INT NOT NULL DEFAULT 0,
|
|
items_removed INT NOT NULL DEFAULT 0,
|
|
error TEXT NOT NULL DEFAULT ''
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS sync_runs_started_idx ON sync_runs (started_at DESC);
|
|
|
|
-- Small key/value store for operator switches (currently just maintenance mode). Kept in
|
|
-- Postgres rather than memory so a restart cannot silently bring the app back up.
|
|
CREATE TABLE IF NOT EXISTS app_settings (
|
|
key TEXT PRIMARY KEY,
|
|
value JSONB NOT NULL,
|
|
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
|
);
|
|
|
|
-- Row-level engagement. One row per reported event; aggregation happens at read time,
|
|
-- which is fine at household scale and keeps the write path trivial.
|
|
CREATE TABLE IF NOT EXISTS row_events (
|
|
id BIGSERIAL PRIMARY KEY,
|
|
occurred_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
emby_user_id TEXT NOT NULL,
|
|
row_id TEXT NOT NULL,
|
|
row_kind TEXT NOT NULL DEFAULT '',
|
|
event TEXT NOT NULL, -- impression | focus | select
|
|
item_id TEXT NOT NULL DEFAULT '',
|
|
dwell_ms INT NOT NULL DEFAULT 0
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS row_events_time_idx ON row_events (occurred_at DESC);
|
|
CREATE INDEX IF NOT EXISTS row_events_row_idx ON row_events (row_id, occurred_at DESC);
|
|
|
|
-- Search terms are retained separately from row engagement so they can inform future
|
|
-- ranking/recommendation work without coupling that analysis to rendered rows.
|
|
CREATE TABLE IF NOT EXISTS search_history (
|
|
id BIGSERIAL PRIMARY KEY,
|
|
occurred_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
emby_user_id TEXT NOT NULL,
|
|
query TEXT NOT NULL
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS search_history_user_time_idx
|
|
ON search_history (emby_user_id, occurred_at DESC);
|
|
|
|
-- Recommendation-relevant Tracearr history. The public Tracearr API has no user or
|
|
-- since cursor, so stable source ids make these rows the durable deduplication boundary.
|
|
-- Deliberately omit artwork, stream-detail blobs and other fields unused by ranking.
|
|
CREATE TABLE IF NOT EXISTS tracearr_sessions (
|
|
server_id TEXT NOT NULL DEFAULT '',
|
|
tracearr_session_id TEXT NOT NULL,
|
|
tracearr_user_id TEXT NOT NULL DEFAULT '',
|
|
username TEXT NOT NULL DEFAULT '',
|
|
state TEXT NOT NULL DEFAULT '',
|
|
media_type TEXT NOT NULL DEFAULT '',
|
|
media_title TEXT NOT NULL DEFAULT '',
|
|
show_title TEXT NOT NULL DEFAULT '',
|
|
season_number INT,
|
|
episode_number INT,
|
|
production_year INT,
|
|
started_at TIMESTAMPTZ,
|
|
stopped_at TIMESTAMPTZ,
|
|
duration_ms BIGINT NOT NULL DEFAULT 0,
|
|
progress_ms BIGINT NOT NULL DEFAULT 0,
|
|
total_duration_ms BIGINT NOT NULL DEFAULT 0,
|
|
watched BOOLEAN NOT NULL DEFAULT false,
|
|
device TEXT NOT NULL DEFAULT '',
|
|
player TEXT NOT NULL DEFAULT '',
|
|
product TEXT NOT NULL DEFAULT '',
|
|
platform TEXT NOT NULL DEFAULT '',
|
|
is_transcode BOOLEAN NOT NULL DEFAULT false,
|
|
video_decision TEXT NOT NULL DEFAULT '',
|
|
audio_decision TEXT NOT NULL DEFAULT '',
|
|
source_video_codec TEXT NOT NULL DEFAULT '',
|
|
source_audio_codec TEXT NOT NULL DEFAULT '',
|
|
emby_item_id TEXT NOT NULL DEFAULT '',
|
|
emby_series_id TEXT NOT NULL DEFAULT '',
|
|
source_fingerprint BYTEA NOT NULL,
|
|
source_seen_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
imported_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
updated_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
PRIMARY KEY (server_id, tracearr_session_id)
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS tracearr_sessions_user_time_idx
|
|
ON tracearr_sessions (tracearr_user_id, started_at DESC);
|
|
CREATE INDEX IF NOT EXISTS tracearr_sessions_username_time_idx
|
|
ON tracearr_sessions (lower(username), started_at DESC);
|
|
CREATE INDEX IF NOT EXISTS tracearr_sessions_emby_item_idx
|
|
ON tracearr_sessions (emby_item_id) WHERE emby_item_id <> '';
|
|
CREATE INDEX IF NOT EXISTS tracearr_sessions_emby_series_idx
|
|
ON tracearr_sessions (emby_series_id) WHERE emby_series_id <> '';
|
|
|
|
-- One compact derived profile per Emby user. Variable affinity maps stay together as
|
|
-- JSON because the builder reads and replaces the whole profile; no request filters
|
|
-- inside these maps.
|
|
CREATE TABLE IF NOT EXISTS recommendation_user_profiles (
|
|
emby_user_id TEXT PRIMARY KEY,
|
|
tracearr_user_id TEXT NOT NULL DEFAULT '',
|
|
tracearr_username TEXT NOT NULL DEFAULT '',
|
|
source_session_count INT NOT NULL DEFAULT 0,
|
|
mean_completion_ratio REAL NOT NULL DEFAULT 0,
|
|
typical_session_minutes INT NOT NULL DEFAULT 0,
|
|
genre_affinity JSONB NOT NULL DEFAULT '{}'::jsonb,
|
|
title_affinity JSONB NOT NULL DEFAULT '{}'::jsonb,
|
|
studio_affinity JSONB NOT NULL DEFAULT '{}'::jsonb,
|
|
codec_outcomes JSONB NOT NULL DEFAULT '{}'::jsonb,
|
|
signals_through TIMESTAMPTZ,
|
|
built_at TIMESTAMPTZ,
|
|
pool_built_at TIMESTAMPTZ,
|
|
dirty_since TIMESTAMPTZ DEFAULT now(),
|
|
last_error TEXT NOT NULL DEFAULT ''
|
|
);
|
|
|
|
CREATE INDEX IF NOT EXISTS recommendation_profiles_dirty_idx
|
|
ON recommendation_user_profiles (dirty_since)
|
|
WHERE dirty_since IS NOT NULL;
|
|
|
|
-- Every eligible ranked title is retained. At household scale this is only tens of
|
|
-- thousands of compact rows and gives short runtime filters far more headroom than the
|
|
-- old 240-title request pool.
|
|
CREATE TABLE IF NOT EXISTS for_you_candidates (
|
|
emby_user_id TEXT NOT NULL,
|
|
item_id TEXT NOT NULL,
|
|
base_rank INT NOT NULL,
|
|
base_score REAL NOT NULL DEFAULT 0,
|
|
runtime_minutes INT NOT NULL DEFAULT 0,
|
|
affinity_score REAL NOT NULL DEFAULT 0,
|
|
compatibility_score REAL NOT NULL DEFAULT 0,
|
|
compatibility_label TEXT NOT NULL DEFAULT '',
|
|
reason_kind TEXT NOT NULL DEFAULT '',
|
|
reason_genre TEXT NOT NULL DEFAULT '',
|
|
reason_source_session_id TEXT NOT NULL DEFAULT '',
|
|
reason_source_item_id TEXT NOT NULL DEFAULT '',
|
|
reason_source_title TEXT NOT NULL DEFAULT '',
|
|
recommendation_reason TEXT NOT NULL DEFAULT '',
|
|
built_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
|
PRIMARY KEY (emby_user_id, item_id),
|
|
FOREIGN KEY (item_id) REFERENCES library_items(id) ON DELETE CASCADE
|
|
);
|
|
|
|
CREATE UNIQUE INDEX IF NOT EXISTS for_you_candidates_user_rank_idx
|
|
ON for_you_candidates (emby_user_id, base_rank);
|
|
CREATE INDEX IF NOT EXISTS for_you_candidates_user_runtime_rank_idx
|
|
ON for_you_candidates (emby_user_id, runtime_minutes, base_rank);
|