blazer.rb 4.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139
  1. require "csv"
  2. require "yaml"
  3. require "chartkick"
  4. require "safely/core"
  5. require "blazer/version"
  6. require "blazer/data_source"
  7. require "blazer/result"
  8. require "blazer/run_statement"
  9. require "blazer/adapters/base_adapter"
  10. require "blazer/adapters/active_record_adapter"
  11. require "blazer/adapters/elasticsearch_adapter"
  12. require "blazer/engine"
  13. module Blazer
  14. class Error < StandardError; end
  15. class TimeoutNotSupported < Error; end
  16. class << self
  17. attr_accessor :audit
  18. attr_reader :time_zone
  19. attr_accessor :user_name
  20. attr_accessor :user_class
  21. attr_accessor :user_method
  22. attr_accessor :before_action
  23. attr_accessor :from_email
  24. attr_accessor :cache
  25. attr_accessor :transform_statement
  26. attr_accessor :check_schedules
  27. attr_accessor :anomaly_checks
  28. attr_accessor :async
  29. attr_accessor :images
  30. end
  31. self.audit = true
  32. self.user_name = :name
  33. self.check_schedules = ["5 minutes", "1 hour", "1 day"]
  34. self.anomaly_checks = false
  35. self.async = false
  36. self.images = false
  37. TIMEOUT_MESSAGE = "Query timed out :("
  38. TIMEOUT_ERRORS = [
  39. "canceling statement due to statement timeout", # postgres
  40. "canceling statement due to conflict with recovery", # postgres
  41. "cancelled on user's request", # redshift
  42. "canceled on user's request", # redshift
  43. "system requested abort", # redshift
  44. "maximum statement execution time exceeded" # mysql
  45. ]
  46. BELONGS_TO_OPTIONAL = {}
  47. BELONGS_TO_OPTIONAL[:optional] = true if Rails::VERSION::MAJOR >= 5
  48. def self.time_zone=(time_zone)
  49. @time_zone = time_zone.is_a?(ActiveSupport::TimeZone) ? time_zone : ActiveSupport::TimeZone[time_zone.to_s]
  50. end
  51. def self.settings
  52. @settings ||= begin
  53. path = Rails.root.join("config", "blazer.yml").to_s
  54. if File.exist?(path)
  55. YAML.load(ERB.new(File.read(path)).result)
  56. else
  57. {}
  58. end
  59. end
  60. end
  61. def self.data_sources
  62. @data_sources ||= begin
  63. ds = Hash[
  64. settings["data_sources"].map do |id, s|
  65. [id, Blazer::DataSource.new(id, s)]
  66. end
  67. ]
  68. ds.default = ds.values.first
  69. ds
  70. end
  71. end
  72. def self.run_checks(schedule: nil)
  73. checks = Blazer::Check.includes(:query)
  74. checks = checks.where(schedule: schedule) if schedule
  75. checks.find_each do |check|
  76. next if check.state == "disabled"
  77. Safely.safely { run_check(check) }
  78. end
  79. end
  80. def self.run_check(check)
  81. rows = nil
  82. error = nil
  83. tries = 1
  84. ActiveSupport::Notifications.instrument("run_check.blazer", check_id: check.id, query_id: check.query.id, state_was: check.state) do |instrument|
  85. # try 3 times on timeout errors
  86. data_source = data_sources[check.query.data_source]
  87. statement = check.query.statement
  88. Blazer.transform_statement.call(data_source, statement) if Blazer.transform_statement
  89. while tries <= 3
  90. result = data_source.run_statement(statement, refresh_cache: true, check: check, query: check.query)
  91. if result.timed_out?
  92. Rails.logger.info "[blazer timeout] query=#{check.query.name}"
  93. tries += 1
  94. sleep(10)
  95. elsif result.error.to_s.start_with?("PG::ConnectionBad")
  96. data_source.reconnect
  97. Rails.logger.info "[blazer reconnect] query=#{check.query.name}"
  98. tries += 1
  99. sleep(10)
  100. else
  101. break
  102. end
  103. end
  104. check.update_state(result)
  105. # TODO use proper logfmt
  106. Rails.logger.info "[blazer check] query=#{check.query.name} state=#{check.state} rows=#{result.rows.try(:size)} error=#{result.error}"
  107. instrument[:statement] = statement
  108. instrument[:data_source] = data_source
  109. instrument[:state] = check.state
  110. instrument[:rows] = result.rows.try(:size)
  111. instrument[:error] = result.error
  112. instrument[:tries] = tries
  113. end
  114. end
  115. def self.send_failing_checks
  116. emails = {}
  117. Blazer::Check.includes(:query).where(state: ["failing", "error", "timed out", "disabled"]).find_each do |check|
  118. check.split_emails.each do |email|
  119. (emails[email] ||= []) << check
  120. end
  121. end
  122. emails.each do |email, checks|
  123. Blazer::CheckMailer.failing_checks(email, checks).deliver_later
  124. end
  125. end
  126. end