gitlab-org--gitlab-foss/app/workers/stuck_ci_jobs_worker.rb

92 lines
2.7 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
class StuckCiJobsWorker # rubocop:disable Scalability/IdempotentWorker
include ApplicationWorker
include CronjobQueue
2015-10-21 09:29:47 +00:00
feature_category :continuous_integration
worker_resource_boundary :cpu
2019-09-04 18:10:43 +00:00
EXCLUSIVE_LEASE_KEY = 'stuck_ci_builds_worker_lease'
2017-02-09 02:29:38 +00:00
BUILD_RUNNING_OUTDATED_TIMEOUT = 1.hour
BUILD_PENDING_OUTDATED_TIMEOUT = 1.day
BUILD_SCHEDULED_OUTDATED_TIMEOUT = 1.hour
BUILD_PENDING_STUCK_TIMEOUT = 1.hour
2015-10-21 09:29:47 +00:00
def perform
return unless try_obtain_lease
Gitlab::AppLogger.info "#{self.class}: Cleaning stuck builds"
2015-10-21 09:29:47 +00:00
drop :running, BUILD_RUNNING_OUTDATED_TIMEOUT, 'ci_builds.updated_at < ?', :stuck_or_timeout_failure
drop :pending, BUILD_PENDING_OUTDATED_TIMEOUT, 'ci_builds.updated_at < ?', :stuck_or_timeout_failure
drop :scheduled, BUILD_SCHEDULED_OUTDATED_TIMEOUT, 'scheduled_at IS NOT NULL AND scheduled_at < ?', :stale_schedule
drop_stuck :pending, BUILD_PENDING_STUCK_TIMEOUT, 'ci_builds.updated_at < ?', :stuck_or_timeout_failure
2017-02-09 02:29:38 +00:00
remove_lease
end
private
def try_obtain_lease
2017-03-01 12:56:54 +00:00
@uuid = Gitlab::ExclusiveLease.new(EXCLUSIVE_LEASE_KEY, timeout: 30.minutes).try_obtain
end
2017-02-09 02:29:38 +00:00
def remove_lease
Gitlab::ExclusiveLease.cancel(EXCLUSIVE_LEASE_KEY, @uuid)
end
def drop(status, timeout, condition, reason)
search(status, timeout, condition) do |build|
drop_build :outdated, build, status, timeout, reason
end
end
def drop_stuck(status, timeout, condition, reason)
search(status, timeout, condition) do |build|
break unless build.stuck?
drop_build :stuck, build, status, timeout, reason
2015-10-21 09:29:47 +00:00
end
end
# rubocop: disable CodeReuse/ActiveRecord
def search(status, timeout, condition)
2017-11-21 18:49:10 +00:00
loop do
jobs = Ci::Build.where(status: status)
.where(condition, timeout.ago)
.includes(:tags, :runner, project: [:namespace, :route])
2017-11-21 18:49:10 +00:00
.limit(100)
.to_a
break if jobs.empty?
jobs.each do |job|
with_context(project: job.project) { yield(job) }
2017-11-21 17:44:52 +00:00
end
end
end
2018-09-27 05:02:24 +00:00
# rubocop: enable CodeReuse/ActiveRecord
def drop_build(type, build, status, timeout, reason)
Gitlab::AppLogger.info "#{self.class}: Dropping #{type} build #{build.id} for runner #{build.runner_id} (status: #{status}, timeout: #{timeout}, reason: #{reason})"
Gitlab::OptimisticLocking.retry_lock(build, 3, name: 'stuck_ci_jobs_worker_drop_build') do |b|
2018-09-27 05:02:24 +00:00
b.drop(reason)
end
rescue StandardError => ex
build.doom!
track_exception_for_build(ex, build)
end
def track_exception_for_build(ex, build)
Gitlab::ErrorTracking.track_exception(ex,
build_id: build.id,
build_name: build.name,
build_stage: build.stage,
pipeline_id: build.pipeline_id,
project_id: build.project_id
)
2018-09-27 05:02:24 +00:00
end
2015-10-21 09:29:47 +00:00
end