Template of a custom feed generator service for the Bluesky network in Ruby
1require 'json'
2require 'sinatra/activerecord'
3require 'skyfall'
4
5require_relative 'config'
6require_relative 'models/feed_post'
7require_relative 'models/post'
8require_relative 'models/subscription'
9
10class FirehoseStream
11 attr_accessor :show_progress, :log_status, :log_posts, :save_posts, :replay_events
12
13 DEFAULT_RELAY = 'bsky.network'
14
15 def initialize(service = nil)
16 @env = (ENV['APP_ENV'] || ENV['RACK_ENV'] || :development).to_sym
17 @service = service || DEFAULT_RELAY
18
19 @show_progress = (@env == :development) ? true : false
20 @log_status = true
21 @log_posts = (@env == :development) ? :matching : false
22 @save_posts = (@env == :development) ? :all : :matching
23 @replay_events = (@env == :development) ? false : true
24
25 @feeds = BlueFactory.all_feeds.select(&:is_updating?)
26 end
27
28 def start
29 return if @sky
30
31 last_cursor = load_or_init_cursor
32 cursor = @replay_events ? last_cursor : nil
33
34 @sky = sky = Skyfall::Firehose.new(@service, :subscribe_repos, cursor)
35
36 @sky.on_message do |m|
37 process_message(m)
38 end
39
40 if @log_status
41 @sky.on_connecting { |u| log "Connecting to #{u}..." }
42 @sky.on_connect {
43 @replaying = !!(cursor)
44 log "Connected ✓"
45 }
46 @sky.on_disconnect {
47 log "Disconnected."
48 save_cursor(sky.cursor)
49 }
50 @sky.on_reconnect { log "Connection lost, reconnecting..." }
51 @sky.on_error { |e| log "ERROR: #{e.class} #{e.message}" }
52 end
53
54 @sky.connect
55 end
56
57 def stop
58 @sky&.disconnect
59 @sky = nil
60 end
61
62 def load_or_init_cursor
63 if sub = Subscription.find_by(service: @service)
64 sub.cursor
65 else
66 Subscription.create!(service: @service, cursor: 0)
67 nil
68 end
69 end
70
71 def save_cursor(cursor)
72 Subscription.where(service: @service).update_all(cursor: cursor)
73 end
74
75 def process_message(msg)
76 if msg.type == :info
77 # AtProto error, the only one right now is "OutdatedCursor"
78 log "InfoMessage: #{msg}"
79
80 elsif msg.type == :identity
81 # use these events if you want to track handle changes:
82 # log "Handle change: #{msg.repo} => #{msg.handle}"
83
84 elsif msg.is_a?(Skyfall::Firehose::UnknownMessage)
85 log "Unknown message type: #{msg.type} (#{msg.seq})"
86 end
87
88 return unless msg.type == :commit
89
90 if @replaying
91 log "Replaying events since #{msg.time.getlocal} -->"
92 @replaying = false
93 end
94
95 if msg.seq % 10 == 0
96 save_cursor(msg.seq)
97 end
98
99 msg.operations.each do |op|
100 case op.type
101 when :bsky_post
102 process_post(msg, op)
103
104 when :bsky_like, :bsky_repost
105 # if you want to use the number of likes and/or reposts for filtering or sorting:
106 # add a likes/reposts column to feeds, then do +1 / -1 here depending on op.action
107
108 when :bsky_follow
109 # if you want to make a personalized feed that needs info about given user's follows/followers:
110 # add a followers table, then add/remove records here depending on op.action
111
112 else
113 # other types like :bsky_block, :bsky_profile (includes profile edits)
114 end
115 end
116 end
117
118 def process_post(msg, op)
119 if op.action == :delete
120 if post = Post.find_by(repo: op.repo, rkey: op.rkey)
121 post.destroy
122 end
123 end
124
125 return unless op.action == :create
126
127 begin
128 if op.raw_record.nil?
129 log "Error: missing expected record data in operation: #{op.uri} (#{msg.seq})"
130 return
131 end
132 rescue CBOR::UnpackError => e
133 log "Error: couldn't decode record data for #{op.uri} (#{msg.seq}): #{e}"
134 return
135 end
136
137 # ignore posts with past date from Twitter etc. imported using some kind of tool
138 begin
139 post_time = Time.parse(op.raw_record['createdAt'])
140 return if post_time < msg.time - 86400
141 rescue StandardError => e
142 log "Skipping post with invalid timestamp: #{op.raw_record['createdAt'].inspect} (#{op.repo}, #{msg.seq})"
143 return
144 end
145
146 text = op.raw_record['text']
147
148 # to save space, delete redundant post text and type from the saved data JSON
149 trimmed_record = op.raw_record.dup
150 trimmed_record.delete('$type')
151 trimmed_record.delete('text')
152 trimmed_json = JSON.generate(trimmed_record)
153
154 # tip: if you don't need full record data for debugging, delete the data column in posts
155 post = Post.new(
156 repo: op.repo,
157 time: msg.time,
158 text: text,
159 rkey: op.rkey,
160 data: trimmed_json,
161 record: op.raw_record
162 )
163
164 matched = false
165
166 @feeds.each do |feed|
167 if feed.post_matches?(post)
168 FeedPost.create!(feed_id: feed.feed_id, post: post, time: msg.time) unless !@save_posts
169 matched = true
170 end
171 end
172
173 if @log_posts == :all || @log_posts && matched
174 puts
175 puts text
176 end
177
178 post.save! if @save_posts == :all
179
180 print '.' if @show_progress && @log_posts != :all
181 rescue StandardError => e
182 log "Error in #process_post: #{e}"
183
184 unless e.message == "nesting of 100 is too deep"
185 log msg.inspect
186 log e.backtrace.reject { |x| x.include?('/ruby/') }
187 end
188 end
189
190 def log(text)
191 puts if @show_progress
192 puts "[#{Time.now}] #{text}"
193 end
194
195 def inspect
196 vars = instance_variables - [:@feeds, :@timer]
197 values = vars.map { |v| "#{v}=#{instance_variable_get(v).inspect}" }.join(", ")
198 "#<#{self.class}:0x#{object_id} #{values}>"
199 end
200end