From e476dbe25baddcbfd5ee61d27c706e16e7771f68 Mon Sep 17 00:00:00 2001 From: Emilien Devos <4016501+unixfox@users.noreply.github.com> Date: Wed, 26 Jul 2023 15:43:43 +0200 Subject: [PATCH] limit feeds and delete materialized views --- config/migrate-scripts/migrate-db-8bc91ce.sh | 6 ++ config/sql/channel_videos.sql | 8 +-- kubernetes/values.yaml | 60 ++++++++++++++++ src/invidious.cr | 12 ---- src/invidious/config.cr | 2 - src/invidious/jobs/refresh_feeds_job.cr | 75 -------------------- src/invidious/routes/account.cr | 2 - src/invidious/routes/login.cr | 3 - src/invidious/search/processors.cr | 18 ++--- src/invidious/users.cr | 39 +++++----- 10 files changed, 101 insertions(+), 124 deletions(-) create mode 100644 config/migrate-scripts/migrate-db-8bc91ce.sh create mode 100644 kubernetes/values.yaml delete mode 100644 src/invidious/jobs/refresh_feeds_job.cr diff --git a/config/migrate-scripts/migrate-db-8bc91ce.sh b/config/migrate-scripts/migrate-db-8bc91ce.sh new file mode 100644 index 000000000..04388175e --- /dev/null +++ b/config/migrate-scripts/migrate-db-8bc91ce.sh @@ -0,0 +1,6 @@ +CREATE INDEX channel_videos_ucid_published_idx + ON public.channel_videos + USING btree + (ucid COLLATE pg_catalog."default", published); + +DROP INDEX channel_videos_ucid_idx; \ No newline at end of file diff --git a/config/sql/channel_videos.sql b/config/sql/channel_videos.sql index cd4e0ffdb..f2ac4876c 100644 --- a/config/sql/channel_videos.sql +++ b/config/sql/channel_videos.sql @@ -19,12 +19,12 @@ CREATE TABLE IF NOT EXISTS public.channel_videos GRANT ALL ON TABLE public.channel_videos TO current_user; --- Index: public.channel_videos_ucid_idx +-- Index: public.channel_videos_ucid_published_idx --- DROP INDEX public.channel_videos_ucid_idx; +-- DROP INDEX public.channel_videos_ucid_published_idx; -CREATE INDEX IF NOT EXISTS channel_videos_ucid_idx +CREATE INDEX IF NOT EXISTS channel_videos_ucid_published_idx ON public.channel_videos USING btree - (ucid COLLATE pg_catalog."default"); + (ucid COLLATE pg_catalog."default", published); diff --git a/kubernetes/values.yaml b/kubernetes/values.yaml new file mode 100644 index 000000000..17d69b5d7 --- /dev/null +++ b/kubernetes/values.yaml @@ -0,0 +1,60 @@ +name: invidious + +image: + repository: quay.io/invidious/invidious + tag: latest + pullPolicy: Always + +replicaCount: 1 + +autoscaling: + enabled: false + minReplicas: 1 + maxReplicas: 16 + targetCPUUtilizationPercentage: 50 + +service: + type: ClusterIP + port: 3000 + #loadBalancerIP: + +resources: {} + #requests: + # cpu: 100m + # memory: 64Mi + #limits: + # cpu: 800m + # memory: 512Mi + +securityContext: + allowPrivilegeEscalation: false + runAsUser: 1000 + runAsGroup: 1000 + fsGroup: 1000 + +# See https://github.com/bitnami/charts/tree/master/bitnami/postgresql +postgresql: + image: + tag: 13 + auth: + username: kemal + password: kemal + database: invidious + primary: + initdb: + username: kemal + password: kemal + scriptsConfigMap: invidious-postgresql-init + +# Adapted from ../config/config.yml +config: + channel_threads: 1 + db: + user: kemal + password: kemal + host: invidious-postgresql + port: 5432 + dbname: invidious + full_refresh: false + https_only: false + domain: diff --git a/src/invidious.cr b/src/invidious.cr index 3804197e3..961ae8721 100644 --- a/src/invidious.cr +++ b/src/invidious.cr @@ -103,14 +103,6 @@ Kemal.config.extra_options do |parser| exit end end - parser.on("-f THREADS", "--feed-threads=THREADS", "Number of threads for refreshing feeds (default: #{CONFIG.feed_threads})") do |number| - begin - CONFIG.feed_threads = number.to_i - rescue ex - puts "THREADS must be integer" - exit - end - end parser.on("-o OUTPUT", "--output=OUTPUT", "Redirect output (default: #{CONFIG.output})") do |output| CONFIG.output = output end @@ -168,10 +160,6 @@ if CONFIG.channel_threads > 0 Invidious::Jobs.register Invidious::Jobs::RefreshChannelsJob.new(PG_DB) end -if CONFIG.feed_threads > 0 - Invidious::Jobs.register Invidious::Jobs::RefreshFeedsJob.new(PG_DB) -end - if CONFIG.statistics_enabled Invidious::Jobs.register Invidious::Jobs::StatisticsRefreshJob.new(PG_DB, SOFTWARE) end diff --git a/src/invidious/config.cr b/src/invidious/config.cr index c4ddcdb36..2e6df47a1 100644 --- a/src/invidious/config.cr +++ b/src/invidious/config.cr @@ -62,8 +62,6 @@ class Config # Time interval between two executions of the job that crawls channel videos (subscriptions update). @[YAML::Field(converter: Preferences::TimeSpanConverter)] property channel_refresh_interval : Time::Span = 30.minutes - # Number of threads to use for updating feeds - property feed_threads : Int32 = 1 # Log file path or STDOUT property output : String = "STDOUT" # Default log level, valid YAML values are ints and strings, see src/invidious/helpers/logger.cr diff --git a/src/invidious/jobs/refresh_feeds_job.cr b/src/invidious/jobs/refresh_feeds_job.cr deleted file mode 100644 index 4f8130dfd..000000000 --- a/src/invidious/jobs/refresh_feeds_job.cr +++ /dev/null @@ -1,75 +0,0 @@ -class Invidious::Jobs::RefreshFeedsJob < Invidious::Jobs::BaseJob - private getter db : DB::Database - - def initialize(@db) - end - - def begin - max_fibers = CONFIG.feed_threads - active_fibers = 0 - active_channel = ::Channel(Bool).new - - loop do - db.query("SELECT email FROM users WHERE feed_needs_update = true OR feed_needs_update IS NULL") do |rs| - rs.each do - email = rs.read(String) - view_name = "subscriptions_#{sha256(email)}" - - if active_fibers >= max_fibers - if active_channel.receive - active_fibers -= 1 - end - end - - active_fibers += 1 - spawn do - begin - # Drop outdated views - column_array = Invidious::Database.get_column_array(db, view_name) - ChannelVideo.type_array.each_with_index do |name, i| - if name != column_array[i]? - LOGGER.info("RefreshFeedsJob: DROP MATERIALIZED VIEW #{view_name}") - db.exec("DROP MATERIALIZED VIEW #{view_name}") - raise "view does not exist" - end - end - - if !db.query_one("SELECT pg_get_viewdef('#{view_name}')", as: String).includes? "WHERE ((cv.ucid = ANY (u.subscriptions))" - LOGGER.info("RefreshFeedsJob: Materialized view #{view_name} is out-of-date, recreating...") - db.exec("DROP MATERIALIZED VIEW #{view_name}") - end - - db.exec("REFRESH MATERIALIZED VIEW #{view_name}") - db.exec("UPDATE users SET feed_needs_update = false WHERE email = $1", email) - rescue ex - # Rename old views - begin - legacy_view_name = "subscriptions_#{sha256(email)[0..7]}" - - db.exec("SELECT * FROM #{legacy_view_name} LIMIT 0") - LOGGER.info("RefreshFeedsJob: RENAME MATERIALIZED VIEW #{legacy_view_name}") - db.exec("ALTER MATERIALIZED VIEW #{legacy_view_name} RENAME TO #{view_name}") - rescue ex - begin - # While iterating through, we may have an email stored from a deleted account - if db.query_one?("SELECT true FROM users WHERE email = $1", email, as: Bool) - LOGGER.info("RefreshFeedsJob: CREATE #{view_name}") - db.exec("CREATE MATERIALIZED VIEW #{view_name} AS #{MATERIALIZED_VIEW_SQL.call(email)}") - db.exec("UPDATE users SET feed_needs_update = false WHERE email = $1", email) - end - rescue ex - LOGGER.error("RefreshFeedJobs: REFRESH #{email} : #{ex.message}") - end - end - end - - active_channel.send(true) - end - end - end - - sleep 5.seconds - Fiber.yield - end - end -end diff --git a/src/invidious/routes/account.cr b/src/invidious/routes/account.cr index dd65e7a60..8086a54e5 100644 --- a/src/invidious/routes/account.cr +++ b/src/invidious/routes/account.cr @@ -123,10 +123,8 @@ module Invidious::Routes::Account return error_template(400, ex) end - view_name = "subscriptions_#{sha256(user.email)}" Invidious::Database::Users.delete(user) Invidious::Database::SessionIDs.delete(email: user.email) - PG_DB.exec("DROP MATERIALIZED VIEW #{view_name}") env.request.cookies.each do |cookie| cookie.expires = Time.utc(1990, 1, 1) diff --git a/src/invidious/routes/login.cr b/src/invidious/routes/login.cr index d0f7ac229..add9f75d3 100644 --- a/src/invidious/routes/login.cr +++ b/src/invidious/routes/login.cr @@ -160,9 +160,6 @@ module Invidious::Routes::Login Invidious::Database::Users.insert(user) Invidious::Database::SessionIDs.insert(sid, email) - view_name = "subscriptions_#{sha256(user.email)}" - PG_DB.exec("CREATE MATERIALIZED VIEW #{view_name} AS #{MATERIALIZED_VIEW_SQL.call(user.email)}") - env.response.cookies["SID"] = Invidious::User::Cookies.sid(CONFIG.domain, sid) if env.request.cookies["PREFS"]? diff --git a/src/invidious/search/processors.cr b/src/invidious/search/processors.cr index 25edb9362..10b81c59e 100644 --- a/src/invidious/search/processors.cr +++ b/src/invidious/search/processors.cr @@ -37,18 +37,18 @@ module Invidious::Search # Search inside of user subscriptions def subscriptions(query : Query, user : Invidious::User) : Array(ChannelVideo) - view_name = "subscriptions_#{sha256(user.email)}" - return PG_DB.query_all(" SELECT id,title,published,updated,ucid,author,length_seconds FROM ( - SELECT *, - to_tsvector(#{view_name}.title) || - to_tsvector(#{view_name}.author) - as document - FROM #{view_name} - ) v_search WHERE v_search.document @@ plainto_tsquery($1) LIMIT 20 OFFSET $2;", - query.text, (query.page - 1) * 20, + SELECT cv.*, + to_tsvector(cv.title) || + to_tsvector(cv.author) AS document + FROM channel_videos cv + JOIN users ON cv.ucid = any(users.subscriptions) + WHERE users.email = $1 AND published > now() - interval '1 month' + ORDER BY published + ) v_search WHERE v_search.document @@ plainto_tsquery($2) LIMIT 20 OFFSET $3;", + user.email, query.text, (query.page - 1) * 20, as: ChannelVideo ) end diff --git a/src/invidious/users.cr b/src/invidious/users.cr index 65566d207..0b2d1ef5f 100644 --- a/src/invidious/users.cr +++ b/src/invidious/users.cr @@ -27,7 +27,6 @@ def get_subscription_feed(user, max_results = 40, page = 1) offset = (page - 1) * limit notifications = Invidious::Database::Users.select_notifications(user) - view_name = "subscriptions_#{sha256(user.email)}" if user.preferences.notifications_only && !notifications.empty? # Only show notifications @@ -53,33 +52,39 @@ def get_subscription_feed(user, max_results = 40, page = 1) # Show latest video from a channel that a user hasn't watched # "unseen_only" isn't really correct here, more accurate would be "unwatched_only" - if user.watched.empty? - values = "'{}'" - else - values = "VALUES #{user.watched.map { |id| %(('#{id}')) }.join(",")}" - end - videos = PG_DB.query_all("SELECT DISTINCT ON (ucid) * FROM #{view_name} WHERE NOT id = ANY (#{values}) ORDER BY ucid, published DESC", as: ChannelVideo) + # "SELECT cv.* FROM channel_videos cv JOIN users ON cv.ucid = any(users.subscriptions) WHERE users.email = $1 AND published > now() - interval '1 month' ORDER BY published DESC" + # "SELECT DISTINCT ON (cv.ucid) cv.* FROM channel_videos cv JOIN users ON cv.ucid = any(users.subscriptions) WHERE users.email = ? AND NOT cv.id = any(users.watched) AND published > now() - interval '1 month' ORDER BY ucid, published DESC" + videos = PG_DB.query_all("SELECT DISTINCT ON (cv.ucid) cv.* " \ + "FROM channel_videos cv " \ + "JOIN users ON cv.ucid = any(users.subscriptions) " \ + "WHERE users.email = $1 AND NOT cv.id = any(users.watched) AND published > now() - interval '1 month' " \ + "ORDER BY ucid, published DESC", user.email, as: ChannelVideo) else # Show latest video from each channel - videos = PG_DB.query_all("SELECT DISTINCT ON (ucid) * FROM #{view_name} ORDER BY ucid, published DESC", as: ChannelVideo) + videos = PG_DB.query_all("SELECT DISTINCT ON (cv.ucid) cv.* " \ + "FROM channel_videos cv " \ + "JOIN users ON cv.ucid = any(users.subscriptions) " \ + "WHERE users.email = $1 AND published > now() - interval '1 month' " \ + "ORDER BY ucid, published DESC", user.email, as: ChannelVideo) end videos.sort_by!(&.published).reverse! else if user.preferences.unseen_only # Only show unwatched - - if user.watched.empty? - values = "'{}'" - else - values = "VALUES #{user.watched.map { |id| %(('#{id}')) }.join(",")}" - end - videos = PG_DB.query_all("SELECT * FROM #{view_name} WHERE NOT id = ANY (#{values}) ORDER BY published DESC LIMIT $1 OFFSET $2", limit, offset, as: ChannelVideo) + videos = PG_DB.query_all("SELECT cv.* " \ + "FROM channel_videos cv " \ + "JOIN users ON cv.ucid = any(users.subscriptions) " \ + "WHERE users.email = $1 AND NOT cv.id = any(users.watched) AND published > now() - interval '1 month' " \ + "ORDER BY published DESC LIMIT $2 OFFSET $3", user.email, limit, offset, as: ChannelVideo) else # Sort subscriptions as normal - - videos = PG_DB.query_all("SELECT * FROM #{view_name} ORDER BY published DESC LIMIT $1 OFFSET $2", limit, offset, as: ChannelVideo) + videos = PG_DB.query_all("SELECT cv.* " \ + "FROM channel_videos cv " \ + "JOIN users ON cv.ucid = any(users.subscriptions) " \ + "WHERE users.email = $1 AND published > now() - interval '1 month' " \ + "ORDER BY published DESC LIMIT $2 OFFSET $3", user.email, limit, offset, as: ChannelVideo) end end