Kieran
4dd9d837a3
* Updated package.json (and made an excuse to make a branch) * Video filepath parser (#6) * Restructured files; Added parser placeholder * More restructuring * Added basic parser for hydrating template strings * Improved docs * More docs * Initial implementation of media profiles (#7) * [WIP] Added basic video download method * [WIP] Very-WIP first steps at parsing options and downloading * Made my options safe by default and removed special safe versions * Ran html generator for mediaprofile model - leaving as-is for now * Addressed a bunch of TODO comments * Add "channel" type Media Source (#8) * [WIP] Working on fetching channel metadata in yt-dlp backend * Finished first draft of methods to do with querying channels * Renamed CommandRunnerMock to have a more descriptive name * Ran the phx generator for the channel model * Renamed Downloader namespace to MediaClient * [WIP] saving before attempting LiveView * LiveView did not work out but here's a working controller how about * Index a channel (#9) * Ran a MediaItem generator; Reformatted to my liking * [WIP] added basic index function * setup oban * Added basic Oban job for indexing * Added in workers for indexing; hooked them into record creation flow * Added a task model with a phx generator * Tied together tasks with jobs and channels * Download indexed videos (#10) * Clarified documentation * more comments * [WIP] hooked up basic video downloading; starting work on metadata * Added metadata model and parsing Adding the metadata model made me realize that, in many cases, yt-dlp returns undesired input in stdout, breaking parsing. In order to get the metadata model working, I had to change the way in which the app interacts with yt-dlp. Now, output is written as a file to disk which is immediately re-read and returned. * Added tests for video download worker * Hooked up video downloading to the channel indexing pipeline * Adds tasks for media items * Updated video metadata parser to extract the title * Ran linting
81 lines
2.8 KiB
Elixir
81 lines
2.8 KiB
Elixir
defmodule Pinchflat.Workers.MediaIndexingWorker do
|
|
@moduledoc false
|
|
|
|
use Oban.Worker,
|
|
queue: :media_indexing,
|
|
unique: [period: :infinity, states: [:available, :scheduled, :retryable]],
|
|
tags: ["media_source", "media_indexing"]
|
|
|
|
alias __MODULE__
|
|
alias Pinchflat.Media
|
|
alias Pinchflat.Tasks
|
|
alias Pinchflat.MediaSource
|
|
alias Pinchflat.Workers.VideoDownloadWorker
|
|
|
|
@impl Oban.Worker
|
|
@doc """
|
|
The ID is that of a channel _record_, not a YouTube channel ID. Indexes
|
|
the provided channel, kicks off downloads for each new MediaItem, and
|
|
reschedules the job to run again in the future (as determined by the
|
|
channel's `index_frequency_minutes` field).
|
|
|
|
README: Re-scheduling here works a little different than you may expect.
|
|
The reschedule time is relative to the time the job has actually _completed_.
|
|
This has some benefits but also side effects to be aware of:
|
|
|
|
- Benefit: No chance for jobs to overlap if a job takes longer than the
|
|
scheduled interval. Less likely to hit API rate limits.
|
|
- Side effect: Intervals are "soft" and _always_ walk forward. This may cause
|
|
user confusion since a 30-minute job scheduled for every hour will
|
|
actually run every 1 hour and 30 minutes. The tradeoff of not inundating
|
|
the API with requests and also not overlapping jobs is worth it, IMO.
|
|
|
|
NOTE: Since indexing can take a LONG time, I should check what happens if an
|
|
application restart occurs while a job is running. Will the job be lost?
|
|
|
|
IDEA: Should I use paging and do indexing in chunks? Is that even faster?
|
|
|
|
Returns :ok | {:ok, %Task{}}
|
|
"""
|
|
def perform(%Oban.Job{args: %{"id" => channel_id}}) do
|
|
channel = MediaSource.get_channel!(channel_id)
|
|
|
|
if channel.index_frequency_minutes <= 0 do
|
|
:ok
|
|
else
|
|
index_media_and_reschedule(channel)
|
|
end
|
|
end
|
|
|
|
defp index_media_and_reschedule(channel) do
|
|
MediaSource.index_media_items(channel)
|
|
enqueue_video_downloads(channel)
|
|
|
|
channel
|
|
|> Map.take([:id])
|
|
|> MediaIndexingWorker.new(schedule_in: channel.index_frequency_minutes * 60)
|
|
|> Tasks.create_job_with_task(channel)
|
|
|> case do
|
|
{:ok, task} -> {:ok, task}
|
|
{:error, :duplicate_job} -> {:ok, :job_exists}
|
|
end
|
|
end
|
|
|
|
# NOTE: this starts a download for each media item that is pending,
|
|
# not just the ones that were indexed in this job run. This should ensure
|
|
# that any stragglers are caught if, for some reason, they weren't enqueued
|
|
# or somehow got de-queued.
|
|
#
|
|
# I'm not sure of a case where this would happen, but it's cheap insurance.
|
|
defp enqueue_video_downloads(channel) do
|
|
channel
|
|
|> Media.list_pending_media_items_for()
|
|
|> Enum.each(fn media_item ->
|
|
media_item
|
|
|> Map.take([:id])
|
|
|> VideoDownloadWorker.new()
|
|
|> Tasks.create_job_with_task(media_item)
|
|
end)
|
|
end
|
|
end
|