Sha256: 983ccf985a8b20f4e1cf81541f04582b77119386695519cfd4905db5c7636009
Contents?: true
Size: 1.69 KB
Versions: 4
Compression:
Stored size: 1.69 KB
Contents
module Workhorse::Jobs class DetectStaleJobsJob # Instantiates a new stale detection job. # # @param locked_to_started_threshold [Integer] The maximum number of seconds # a job is allowed to stay 'locked' before this job throws an exception. # Set this to 0 to skip this check. # @param run_time_threshold [Integer] The maximum number of seconds # a job is allowed to run before this job throws an exception. Set this to # 0 to skip this check. def initialize(locked_to_started_threshold: 3 * 60, run_time_threshold: 12 * 60) @locked_to_started_threshold = locked_to_started_threshold @run_time_threshold = run_time_threshold end def perform messages = [] # Detect jobs that are locked for too long # if @locked_to_started_threshold != 0 rel = Workhorse::DbJob.locked rel = rel.where('locked_at < ?', @locked_to_started_threshold.seconds.ago) ids = rel.pluck(:id) if ids.size > 0 messages << "Detected #{ids.size} jobs that were locked more than "\ "#{@locked_to_started_threshold}s ago and might be stale: #{ids.inspect}." end end # Detect jobs that are running for too long # if @run_time_threshold != 0 rel = Workhorse::DbJob.started rel = rel.where('started_at < ?', @run_time_threshold.seconds.ago) ids = rel.pluck(:id) if ids.size > 0 messages << "Detected #{ids.size} jobs that are running for longer than "\ "#{@run_time_threshold}s ago and might be stale: #{ids.inspect}." end end if messages.any? fail messages.join(' ') end end end end
Version data entries
4 entries across 4 versions & 1 rubygems