blazer.rb 4.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135
  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/engine"
  9. module Blazer
  10. class Error < StandardError; end
  11. class TimeoutNotSupported < Error; end
  12. class << self
  13. attr_accessor :audit
  14. attr_reader :time_zone
  15. attr_accessor :user_name
  16. attr_accessor :user_class
  17. attr_accessor :user_method
  18. attr_accessor :before_action
  19. attr_accessor :from_email
  20. attr_accessor :cache
  21. attr_accessor :transform_statement
  22. attr_accessor :check_schedules
  23. attr_accessor :anomaly_checks
  24. attr_accessor :async
  25. attr_accessor :images
  26. end
  27. self.audit = true
  28. self.user_name = :name
  29. self.check_schedules = ["5 minutes", "1 hour", "1 day"]
  30. self.anomaly_checks = false
  31. self.async = false
  32. self.images = false
  33. TIMEOUT_MESSAGE = "Query timed out :("
  34. TIMEOUT_ERRORS = [
  35. "canceling statement due to statement timeout", # postgres
  36. "canceling statement due to conflict with recovery", # postgres
  37. "cancelled on user's request", # redshift
  38. "canceled on user's request", # redshift
  39. "system requested abort", # redshift
  40. "maximum statement execution time exceeded" # mysql
  41. ]
  42. BELONGS_TO_OPTIONAL = {}
  43. BELONGS_TO_OPTIONAL[:optional] = true if Rails::VERSION::MAJOR >= 5
  44. def self.time_zone=(time_zone)
  45. @time_zone = time_zone.is_a?(ActiveSupport::TimeZone) ? time_zone : ActiveSupport::TimeZone[time_zone.to_s]
  46. end
  47. def self.settings
  48. @settings ||= begin
  49. path = Rails.root.join("config", "blazer.yml").to_s
  50. if File.exist?(path)
  51. YAML.load(ERB.new(File.read(path)).result)
  52. else
  53. {}
  54. end
  55. end
  56. end
  57. def self.data_sources
  58. @data_sources ||= begin
  59. ds = Hash[
  60. settings["data_sources"].map do |id, s|
  61. [id, Blazer::DataSource.new(id, s)]
  62. end
  63. ]
  64. ds.default = ds.values.first
  65. ds
  66. end
  67. end
  68. def self.run_checks(schedule: nil)
  69. checks = Blazer::Check.includes(:query)
  70. checks = checks.where(schedule: schedule) if schedule
  71. checks.find_each do |check|
  72. next if check.state == "disabled"
  73. Safely.safely { run_check(check) }
  74. end
  75. end
  76. def self.run_check(check)
  77. rows = nil
  78. error = nil
  79. tries = 1
  80. ActiveSupport::Notifications.instrument("run_check.blazer", check_id: check.id, query_id: check.query.id, state_was: check.state) do |instrument|
  81. # try 3 times on timeout errors
  82. data_source = data_sources[check.query.data_source]
  83. statement = check.query.statement
  84. Blazer.transform_statement.call(data_source, statement) if Blazer.transform_statement
  85. while tries <= 3
  86. result = data_source.run_statement(statement, refresh_cache: true, check: check, query: check.query)
  87. if result.timed_out?
  88. Rails.logger.info "[blazer timeout] query=#{check.query.name}"
  89. tries += 1
  90. sleep(10)
  91. elsif result.error.to_s.start_with?("PG::ConnectionBad")
  92. data_source.reconnect
  93. Rails.logger.info "[blazer reconnect] query=#{check.query.name}"
  94. tries += 1
  95. sleep(10)
  96. else
  97. break
  98. end
  99. end
  100. check.update_state(result)
  101. # TODO use proper logfmt
  102. Rails.logger.info "[blazer check] query=#{check.query.name} state=#{check.state} rows=#{result.rows.try(:size)} error=#{result.error}"
  103. instrument[:statement] = statement
  104. instrument[:data_source] = data_source
  105. instrument[:state] = check.state
  106. instrument[:rows] = result.rows.try(:size)
  107. instrument[:error] = result.error
  108. instrument[:tries] = tries
  109. end
  110. end
  111. def self.send_failing_checks
  112. emails = {}
  113. Blazer::Check.includes(:query).where(state: ["failing", "error", "timed out", "disabled"]).find_each do |check|
  114. check.split_emails.each do |email|
  115. (emails[email] ||= []) << check
  116. end
  117. end
  118. emails.each do |email, checks|
  119. Blazer::CheckMailer.failing_checks(email, checks).deliver_later
  120. end
  121. end
  122. end