Files
jellyfin-scene-search/scene_search/meili.py
T

237 lines
7.8 KiB
Python

from __future__ import annotations
import json
import time
from collections.abc import Iterable
from typing import Callable
import requests
class Meili:
def __init__(self, url: str, master_key: str, index: str) -> None:
self.url = url.rstrip("/")
self.index = index
self.session = requests.Session()
if master_key:
self.session.headers.update({"Authorization": f"Bearer {master_key}"})
self.session.headers.update({"Content-Type": "application/json"})
def _check(self, response: requests.Response) -> requests.Response:
if not response.ok:
raise RuntimeError(f"Meilisearch {response.status_code}: {response.text}")
return response
@staticmethod
def _json_vector(vector: Iterable[float]) -> list[float]:
"""Convert NumPy/PyTorch scalar values into JSON-native floats."""
return [float(value) for value in vector]
def ensure_index(self) -> None:
r = self.session.get(f"{self.url}/indexes/{self.index}")
if r.status_code == 404:
self._check(self.session.post(
f"{self.url}/indexes",
json={"uid": self.index, "primaryKey": "id"},
))
else:
self._check(r)
def configure_user_vectors(self, dimensions: int) -> None:
self.ensure_index()
self._check(self.session.patch(
f"{self.url}/indexes/{self.index}/settings/embedders",
json={"siglip2": {"source": "userProvided", "dimensions": dimensions}},
))
self._check(self.session.put(
f"{self.url}/indexes/{self.index}/settings/filterable-attributes",
json=[
"media_id",
"title",
"caption_model",
"caption_route",
"route_model",
"source_kind",
"jellyfin_item_id",
"jellyfin_library_id",
"jellyfin_item_type",
"jellyfin_tags",
],
))
self._check(self.session.put(
f"{self.url}/indexes/{self.index}/settings/searchable-attributes",
json=["title", "caption", "jellyfin_tags"],
))
def clear_documents(self) -> None:
self.ensure_index()
self._check(self.session.delete(
f"{self.url}/indexes/{self.index}/documents",
))
def add_documents(
self,
documents: list[dict],
*,
wait: bool = False,
) -> int | None:
"""Replace complete documents (used for indexing and targeted repairs)."""
if not documents:
return
response = self._check(self.session.post(
f"{self.url}/indexes/{self.index}/documents",
params={"primaryKey": "id"},
json=documents,
))
task_uid = response.json().get("taskUid")
if task_uid is None:
if wait:
raise RuntimeError("Meilisearch replacement did not return a taskUid")
return None
task_uid = int(task_uid)
if wait:
self.wait_for_task(task_uid)
return task_uid
def get_task(self, task_uid: int) -> dict:
"""Read the durable status of an asynchronous Meilisearch task."""
return self._check(
self.session.get(f"{self.url}/tasks/{task_uid}", timeout=15)
).json()
def wait_for_task(
self,
task_uid: int,
*,
timeout_seconds: float | None = 120,
poll_seconds: float = 2.0,
on_poll: Callable[[dict], None] | None = None,
) -> dict:
"""Wait for a task. None timeout permits an intentionally long repair."""
deadline = (
time.monotonic() + timeout_seconds
if timeout_seconds is not None
else None
)
while True:
task = self.get_task(task_uid)
status = str(task.get("status") or "")
if status == "succeeded":
return task
if status in {"failed", "canceled"}:
error = task.get("error") or {}
message = error.get("message") if isinstance(error, dict) else error
raise RuntimeError(
f"Meilisearch task {task_uid} {status}: "
f"{message or task}"
)
if on_poll is not None:
on_poll(task)
if deadline is not None and time.monotonic() >= deadline:
raise RuntimeError(
f"Timed out waiting for Meilisearch task {task_uid} "
f"(last status: {status or 'unknown'})"
)
time.sleep(max(0.1, poll_seconds))
def update_documents(
self,
documents: list[dict],
*,
wait: bool = False,
) -> None:
"""Update only supplied caption fields, preserving identities and vectors.
POST replaces an entire Meilisearch document; PUT is the partial-update
endpoint. Using POST here previously deleted titles, Jellyfin item IDs,
and SigLIP vectors whenever a caption was synchronised.
"""
if not documents:
return
self.ensure_index()
response = self._check(
self.session.put(
f"{self.url}/indexes/{self.index}/documents",
params={"primaryKey": "id"},
json=documents,
)
)
if wait:
task_uid = response.json().get("taskUid")
if task_uid is None:
raise RuntimeError(
"Meilisearch update did not return a taskUid"
)
self.wait_for_task(int(task_uid))
def delete_jellyfin_libraries(self, library_ids: list[str]) -> None:
if not library_ids:
return
self.ensure_index()
filter_expression = (
"jellyfin_library_id IN " + json.dumps(library_ids)
)
self._check(
self.session.post(
f"{self.url}/indexes/{self.index}/documents/delete",
json={"filter": filter_expression},
)
)
def stats(self) -> dict:
self.ensure_index()
return self._check(
self.session.get(f"{self.url}/indexes/{self.index}/stats")
).json()
def search_caption_text(
self,
query: str,
limit: int = 20,
filter_expression: str | None = None,
) -> dict:
"""Find literal caption matches without visual-vector dilution.
The field restriction prevents title and tag matches from masquerading
as caption hits. Requiring all terms avoids loose one-word matches
displacing ordinary Jellyfin search results.
"""
payload = {
"q": query,
"limit": limit,
"attributesToSearchOn": ["caption"],
"matchingStrategy": "all",
"showRankingScore": True,
}
if filter_expression:
payload["filter"] = filter_expression
return self._check(self.session.post(
f"{self.url}/indexes/{self.index}/search",
json=payload,
)).json()
def search_vector(
self,
vector: Iterable[float],
query: str,
limit: int = 10,
semantic_ratio: float = 0.8,
filter_expression: str | None = None,
) -> dict:
payload = {
"q": query,
"vector": self._json_vector(vector),
"hybrid": {
"embedder": "siglip2",
"semanticRatio": semantic_ratio,
},
"limit": limit,
"showRankingScore": True,
}
if filter_expression:
payload["filter"] = filter_expression
return self._check(self.session.post(
f"{self.url}/indexes/{self.index}/search",
json=payload,
)).json()