sheetsync: Optionally download media linked in image links column

To enable this, you need to:
- set --media-dir globally for sheetsync
- enable download_media=true for the events sync config
To disable for individual rows (eg. because of known issues), put "[nodownload]" in the notes column.
Mike Lang 2 months ago
parent 32f573297d
commit 7734444274

@ -3,6 +3,7 @@ import json
import logging
import signal
from collections import defaultdict
from urllib.parse import urlparse
import argh
import gevent.backdoor
@ -15,6 +16,7 @@ from requests import HTTPError
import common
import common.dateutil
from common.database import DBManager, query, get_column_placeholder
from common.media import check_for_media, download_media
from common.sheets import Sheets as SheetsClient
from .sheets import SheetsEventsMiddleware, SheetsPlaylistsMiddleware, SheetsArchiveMiddleware
@ -274,6 +276,11 @@ class EventsSync(SheetSync):
"category",
}
def __init__(self, name, middleware, stop, dbmanager, reverse_sync=False, media_dir=None):
super().__init__(name, middleware, stop, dbmanager, reverse_sync)
self.media_dir = media_dir
self.media_downloads = None if media_dir is None else {}
def observe_rows(self, rows):
counts = defaultdict(lambda: 0)
for row in rows:
@ -287,6 +294,25 @@ class EventsSync(SheetSync):
def sync_row(self, sheet_row, db_row):
# Do some special-case transforms for events before syncing
# Attempt to download any URLs in the links column if we don't already have them.
# This is done asyncronously. We keep a record of failed attempts for two reasons:
# - To avoid retrying
# - To populate the errors column asyncronously
# This record is just in memory - we're ok retrying after every restart.
# You can disable downloads on a per-row basis by putting "[nodownload]" in the notes column.
if sheet_row is not None and self.media_dir is not None and "[nodownload]" not in sheet_row["notes"]:
for url in sheet_row['image_links']:
if url not in self.media_downloads:
self.media_downloads[url] = gevent.spawn(self.download_media, url)
# Greenlet.exception is populated if the greenlet failed with an exception,
# or None otherwise (success or not finished).
# We treat a failure to fetch a URL like a parse error.
e = self.media_downloads[url].exception
if e is not None:
sheet_row.setdefault("_parse_errors", []).append(
f"Failed to download media link {url:!r}: {e}"
)
if db_row is not None:
# If no database error, but we have parse errors, indicate they should be displayed.
if db_row.error is None and sheet_row is not None and sheet_row.get('_parse_errors'):
@ -300,6 +326,20 @@ class EventsSync(SheetSync):
super().sync_row(sheet_row, db_row)
def download_media(self, url):
hostname = urlparse(url).hostname
if hostname in ("youtu.be", "youtube.com"):
self.logger.info(f"Ignoring url {url:!r}: Blocklisted hostname")
if check_for_media(self.media_dir, url):
self.logger.info(f"Already have content for url {url:!r}")
return
try:
download_media(url, self.media_dir)
except Exception:
self.logger.warning(f"Failed to download url {url:!r}", exc_info=True)
raise
self.logger.info(f"Downloaded media for url {url:!r}")
class ArchiveSync(EventsSync):
# Archive events are a special case of event with less input columns.
@ -372,7 +412,7 @@ class PlaylistsSync(SheetSync):
event_id: The id of the streamlog event to sync
""",
)
def main(dbconnect, sync_configs, metrics_port=8005, backdoor_port=0):
def main(dbconnect, sync_configs, metrics_port=8005, backdoor_port=0, media_dir="."):
"""
Sheet sync constantly scans a Google Sheets sheet and a database, copying inputs from the sheet
to the DB and outputs from the DB to the sheet.
@ -466,7 +506,10 @@ def main(dbconnect, sync_configs, metrics_port=8005, backdoor_port=0):
"playlists": PlaylistsSync,
"archive": ArchiveSync,
}[config["type"]]
sync = sync_class(config["name"], middleware, stop, dbmanager, reverse_sync)
sync_class_kwargs = {}
if config["type"] == "events" and config.get("download_media", False):
sync_class_kwargs["media_dir"] = media_dir
sync = sync_class(config["name"], middleware, stop, dbmanager, reverse_sync, **sync_class_kwargs)
workers.append(sync)
jobs = [gevent.spawn(worker.run) for worker in workers]

Loading…
Cancel
Save