website_agent.rb 14KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383
  1. require 'nokogiri'
  2. require 'date'
  3. module Agents
  4. class WebsiteAgent < Agent
  5. include WebRequestConcern
  6. default_schedule "every_12h"
  7. UNIQUENESS_LOOK_BACK = 200
  8. UNIQUENESS_FACTOR = 3
  9. description <<-MD
  10. The WebsiteAgent scrapes a website, XML document, or JSON feed and creates Events based on the results.
  11. Specify a `url` and select a `mode` for when to create Events based on the scraped data, either `all` or `on_change`.
  12. `url` can be a single url, or an array of urls (for example, for multiple pages with the exact same structure but different content to scrape)
  13. The `type` value can be `xml`, `html`, `json`, or `text`.
  14. To tell the Agent how to parse the content, specify `extract` as a hash with keys naming the extractions and values of hashes.
  15. When parsing HTML or XML, these sub-hashes specify how each extraction should be done. The Agent first selects a node set from the document for each extraction key by evaluating either a CSS selector in `css` or an XPath expression in `xpath`. It then evaluates an XPath expression in `value` on each node in the node set, converting the result into string. Here's an example:
  16. "extract": {
  17. "url": { "css": "#comic img", "value": "@src" },
  18. "title": { "css": "#comic img", "value": "@title" },
  19. "body_text": { "css": "div.main", "value": ".//text()" }
  20. }
  21. "@_attr_" is the XPath expression to extract the value of an attribute named _attr_ from a node, and ".//text()" is to extract all the enclosed texts. You can also use [XPath functions](http://www.w3.org/TR/xpath/#section-String-Functions) like `normalize-space` to strip and squeeze whitespace, `substring-after` to extract part of a text, and `translate` to remove comma from a formatted number, etc. Note that these functions take a string, not a node set, so what you may think would be written as `normalize-space(.//text())` should actually be `normalize-space(.)`.
  22. When parsing JSON, these sub-hashes specify [JSONPaths](http://goessner.net/articles/JsonPath/) to the values that you care about. For example:
  23. "extract": {
  24. "title": { "path": "results.data[*].title" },
  25. "description": { "path": "results.data[*].description" }
  26. }
  27. When parsing text, each sub-hash should contain a `regexp` and `index`. Output text is matched against the regular expression repeatedly from the beginning through to the end, collecting a captured group specified by `index` in each match. Each index should be either an integer or a string name which corresponds to <code>(?&lt;<em>name</em>&gt;...)</code>. For example, to parse lines of <code><em>word</em>: <em>definition</em></code>, the following should work:
  28. "extract": {
  29. "word": { "regexp": "^(.+?): (.+)$", index: 1 },
  30. "definition": { "regexp": "^(.+?): (.+)$", index: 2 }
  31. }
  32. Or if you prefer names to numbers for index:
  33. "extract": {
  34. "word": { "regexp": "^(?<word>.+?): (?<definition>.+)$", index: 'word' },
  35. "definition": { "regexp": "^(?<word>.+?): (?<definition>.+)$", index: 'definition' }
  36. }
  37. To extract the whole content as one event:
  38. "extract": {
  39. "content": { "regexp": "\A(?m:.)*\z", index: 0 }
  40. }
  41. Beware that `.` does not match the newline character (LF) unless the `m` flag is in effect, and `^`/`$` basically match every line beginning/end. See [this document](http://ruby-doc.org/core-#{RUBY_VERSION}/doc/regexp_rdoc.html) to learn the regular expression variant used in this service.
  42. Note that for all of the formats, whatever you extract MUST have the same number of matches for each extractor. E.g., if you're extracting rows, all extractors must match all rows. For generating CSS selectors, something like [SelectorGadget](http://selectorgadget.com) may be helpful.
  43. Can be configured to use HTTP basic auth by including the `basic_auth` parameter with `"username:password"`, or `["username", "password"]`.
  44. Set `expected_update_period_in_days` to the maximum amount of time that you'd expect to pass between Events being created by this Agent. This is only used to set the "working" status.
  45. Set `uniqueness_look_back` to limit the number of events checked for uniqueness (typically for performance). This defaults to the larger of #{UNIQUENESS_LOOK_BACK} or #{UNIQUENESS_FACTOR}x the number of detected received results.
  46. Set `force_encoding` to an encoding name if the website does not return a Content-Type header with a proper charset.
  47. Set `user_agent` to a custom User-Agent name if the website does not like the default value (`#{default_user_agent}`).
  48. The `headers` field is optional. When present, it should be a hash of headers to send with the request.
  49. The WebsiteAgent can also scrape based on incoming events. It will scrape the url contained in the `url` key of the incoming event payload. If you specify `merge` as the mode, it will retain the old payload and update it with the new values.
  50. In Liquid templating, the following variable is available:
  51. * `_response_`: A response object with the following keys:
  52. * `status`: HTTP status as integer. (Almost always 200)
  53. * `headers`: Reponse headers; for example, `{{ _response_.headers.Content-Type }}` expands to the value of the Content-Type header. Keys are insentitive to cases and -/_.
  54. MD
  55. event_description do
  56. "Events will have the following fields:\n\n %s" % [
  57. Utils.pretty_print(Hash[options['extract'].keys.map { |key|
  58. [key, "..."]
  59. }])
  60. ]
  61. end
  62. def working?
  63. event_created_within?(interpolated['expected_update_period_in_days']) && !recent_error_logs?
  64. end
  65. def default_options
  66. {
  67. 'expected_update_period_in_days' => "2",
  68. 'url' => "http://xkcd.com",
  69. 'type' => "html",
  70. 'mode' => "on_change",
  71. 'extract' => {
  72. 'url' => { 'css' => "#comic img", 'value' => "@src" },
  73. 'title' => { 'css' => "#comic img", 'value' => "@alt" },
  74. 'hovertext' => { 'css' => "#comic img", 'value' => "@title" }
  75. }
  76. }
  77. end
  78. def validate_options
  79. # Check for required fields
  80. errors.add(:base, "url and expected_update_period_in_days are required") unless options['expected_update_period_in_days'].present? && options['url'].present?
  81. if !options['extract'].present? && extraction_type != "json"
  82. errors.add(:base, "extract is required for all types except json")
  83. end
  84. # Check for optional fields
  85. if options['mode'].present?
  86. errors.add(:base, "mode must be set to on_change, all or merge") unless %w[on_change all merge].include?(options['mode'])
  87. end
  88. if options['expected_update_period_in_days'].present?
  89. errors.add(:base, "Invalid expected_update_period_in_days format") unless is_positive_integer?(options['expected_update_period_in_days'])
  90. end
  91. if options['uniqueness_look_back'].present?
  92. errors.add(:base, "Invalid uniqueness_look_back format") unless is_positive_integer?(options['uniqueness_look_back'])
  93. end
  94. if (encoding = options['force_encoding']).present?
  95. case encoding
  96. when String
  97. begin
  98. Encoding.find(encoding)
  99. rescue ArgumentError
  100. errors.add(:base, "Unknown encoding: #{encoding.inspect}")
  101. end
  102. else
  103. errors.add(:base, "force_encoding must be a string")
  104. end
  105. end
  106. validate_web_request_options!
  107. end
  108. def check
  109. check_urls(interpolated['url'])
  110. end
  111. def check_urls(in_url)
  112. return unless in_url.present?
  113. Array(in_url).each do |url|
  114. check_url(url)
  115. end
  116. end
  117. def check_url(url, payload = {})
  118. log "Fetching #{url}"
  119. response = faraday.get(url)
  120. raise "Failed: #{response.inspect}" unless response.success?
  121. interpolation_context.stack {
  122. interpolation_context['_response_'] = ResponseDrop.new(response)
  123. body = response.body
  124. if (encoding = interpolated['force_encoding']).present?
  125. body = body.encode(Encoding::UTF_8, encoding)
  126. end
  127. doc = parse(body)
  128. if extract_full_json?
  129. if store_payload!(previous_payloads(1), doc)
  130. log "Storing new result for '#{name}': #{doc.inspect}"
  131. create_event payload: payload.merge(doc)
  132. end
  133. return
  134. end
  135. output =
  136. case extraction_type
  137. when 'json'
  138. extract_json(doc)
  139. when 'text'
  140. extract_text(doc)
  141. else
  142. extract_xml(doc)
  143. end
  144. num_unique_lengths = interpolated['extract'].keys.map { |name| output[name].length }.uniq
  145. if num_unique_lengths.length != 1
  146. raise "Got an uneven number of matches for #{interpolated['name']}: #{interpolated['extract'].inspect}"
  147. end
  148. old_events = previous_payloads num_unique_lengths.first
  149. num_unique_lengths.first.times do |index|
  150. result = {}
  151. interpolated['extract'].keys.each do |name|
  152. result[name] = output[name][index]
  153. if name.to_s == 'url'
  154. result[name] = (response.env[:url] + result[name]).to_s
  155. end
  156. end
  157. if store_payload!(old_events, result)
  158. log "Storing new parsed result for '#{name}': #{result.inspect}"
  159. create_event payload: payload.merge(result)
  160. end
  161. end
  162. }
  163. rescue => e
  164. error "Error when fetching url: #{e.message}\n#{e.backtrace.join("\n")}"
  165. end
  166. def receive(incoming_events)
  167. incoming_events.each do |event|
  168. interpolate_with(event) do
  169. url_to_scrape = event.payload['url']
  170. next unless url_to_scrape =~ /^https?:\/\//i
  171. check_url(url_to_scrape,
  172. interpolated['mode'].to_s == "merge" ? event.payload : {})
  173. end
  174. end
  175. end
  176. private
  177. # This method returns true if the result should be stored as a new event.
  178. # If mode is set to 'on_change', this method may return false and update an existing
  179. # event to expire further in the future.
  180. def store_payload!(old_events, result)
  181. case interpolated['mode'].presence
  182. when 'on_change'
  183. result_json = result.to_json
  184. old_events.each do |old_event|
  185. if old_event.payload.to_json == result_json
  186. old_event.expires_at = new_event_expiration_date
  187. old_event.save!
  188. return false
  189. end
  190. end
  191. true
  192. when 'all', 'merge', ''
  193. true
  194. else
  195. raise "Illegal options[mode]: #{interpolated['mode']}"
  196. end
  197. end
  198. def previous_payloads(num_events)
  199. if interpolated['uniqueness_look_back'].present?
  200. look_back = interpolated['uniqueness_look_back'].to_i
  201. else
  202. # Larger of UNIQUENESS_FACTOR * num_events and UNIQUENESS_LOOK_BACK
  203. look_back = UNIQUENESS_FACTOR * num_events
  204. if look_back < UNIQUENESS_LOOK_BACK
  205. look_back = UNIQUENESS_LOOK_BACK
  206. end
  207. end
  208. events.order("id desc").limit(look_back) if interpolated['mode'] == "on_change"
  209. end
  210. def extract_full_json?
  211. !interpolated['extract'].present? && extraction_type == "json"
  212. end
  213. def extraction_type
  214. (interpolated['type'] || begin
  215. case interpolated['url']
  216. when /\.(rss|xml)$/i
  217. "xml"
  218. when /\.json$/i
  219. "json"
  220. when /\.(txt|text)$/i
  221. "text"
  222. else
  223. "html"
  224. end
  225. end).to_s
  226. end
  227. def extract_each(doc, &block)
  228. interpolated['extract'].each_with_object({}) { |(name, extraction_details), output|
  229. output[name] = block.call(extraction_details)
  230. }
  231. end
  232. def extract_json(doc)
  233. extract_each(doc) { |extraction_details|
  234. result = Utils.values_at(doc, extraction_details['path'])
  235. log "Extracting #{extraction_type} at #{extraction_details['path']}: #{result}"
  236. result
  237. }
  238. end
  239. def extract_text(doc)
  240. extract_each(doc) { |extraction_details|
  241. regexp = Regexp.new(extraction_details['regexp'])
  242. result = []
  243. doc.scan(regexp) {
  244. result << Regexp.last_match[extraction_details['index']]
  245. }
  246. log "Extracting #{extraction_type} at #{regexp}: #{result}"
  247. result
  248. }
  249. end
  250. def extract_xml(doc)
  251. extract_each(doc) { |extraction_details|
  252. case
  253. when css = extraction_details['css']
  254. nodes = doc.css(css)
  255. when xpath = extraction_details['xpath']
  256. doc.remove_namespaces! # ignore xmlns, useful when parsing atom feeds
  257. nodes = doc.xpath(xpath)
  258. else
  259. raise '"css" or "xpath" is required for HTML or XML extraction'
  260. end
  261. case nodes
  262. when Nokogiri::XML::NodeSet
  263. result = nodes.map { |node|
  264. case value = node.xpath(extraction_details['value'])
  265. when Float
  266. # Node#xpath() returns any numeric value as float;
  267. # convert it to integer as appropriate.
  268. value = value.to_i if value.to_i == value
  269. end
  270. value.to_s
  271. }
  272. else
  273. raise "The result of HTML/XML extraction was not a NodeSet"
  274. end
  275. log "Extracting #{extraction_type} at #{xpath || css}: #{result}"
  276. result
  277. }
  278. end
  279. def parse(data)
  280. case extraction_type
  281. when "xml"
  282. Nokogiri::XML(data)
  283. when "json"
  284. JSON.parse(data)
  285. when "html"
  286. Nokogiri::HTML(data)
  287. when "text"
  288. data
  289. else
  290. raise "Unknown extraction type #{extraction_type}"
  291. end
  292. end
  293. def is_positive_integer?(value)
  294. Integer(value) >= 0
  295. rescue
  296. false
  297. end
  298. # Wraps Faraday::Response
  299. class ResponseDrop < LiquidDroppable::Drop
  300. def headers
  301. HeaderDrop.new(@object.headers)
  302. end
  303. # Integer value of HTTP status
  304. def status
  305. @object.status
  306. end
  307. end
  308. # Wraps Faraday::Utilsa::Headers
  309. class HeaderDrop < LiquidDroppable::Drop
  310. def before_method(name)
  311. @object[name.tr('_', '-')]
  312. end
  313. end
  314. end
  315. end