debian-mirror-gitlab/app/workers/stuck_ci_jobs_worker.rb

92 lines
2.7 KiB
Ruby
Raw Normal View History

2018-11-08 19:23:39 +05:30
# frozen_string_literal: true
2020-04-08 14:13:33 +05:30
class StuckCiJobsWorker # rubocop:disable Scalability/IdempotentWorker
2018-03-17 18:26:18 +05:30
include ApplicationWorker
2017-08-17 22:00:37 +05:30
include CronjobQueue
2019-12-21 20:55:43 +05:30
feature_category :continuous_integration
2019-12-26 22:10:19 +05:30
worker_resource_boundary :cpu
2019-12-21 20:55:43 +05:30
2019-12-04 20:38:33 +05:30
EXCLUSIVE_LEASE_KEY = 'stuck_ci_builds_worker_lease'
2017-08-17 22:00:37 +05:30
BUILD_RUNNING_OUTDATED_TIMEOUT = 1.hour
BUILD_PENDING_OUTDATED_TIMEOUT = 1.day
2018-12-05 23:21:45 +05:30
BUILD_SCHEDULED_OUTDATED_TIMEOUT = 1.hour
2017-08-17 22:00:37 +05:30
BUILD_PENDING_STUCK_TIMEOUT = 1.hour
def perform
return unless try_obtain_lease
2021-01-03 14:25:43 +05:30
Gitlab::AppLogger.info "#{self.class}: Cleaning stuck builds"
2017-08-17 22:00:37 +05:30
2018-12-05 23:21:45 +05:30
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
2017-08-17 22:00:37 +05:30
remove_lease
end
private
def try_obtain_lease
@uuid = Gitlab::ExclusiveLease.new(EXCLUSIVE_LEASE_KEY, timeout: 30.minutes).try_obtain
end
def remove_lease
Gitlab::ExclusiveLease.cancel(EXCLUSIVE_LEASE_KEY, @uuid)
end
2018-12-05 23:21:45 +05:30
def drop(status, timeout, condition, reason)
search(status, timeout, condition) do |build|
drop_build :outdated, build, status, timeout, reason
2017-08-17 22:00:37 +05:30
end
end
2018-12-05 23:21:45 +05:30
def drop_stuck(status, timeout, condition, reason)
search(status, timeout, condition) do |build|
2018-10-15 14:42:47 +05:30
break unless build.stuck?
2018-03-17 18:26:18 +05:30
2018-12-05 23:21:45 +05:30
drop_build :stuck, build, status, timeout, reason
2017-08-17 22:00:37 +05:30
end
end
2018-12-05 23:21:45 +05:30
# rubocop: disable CodeReuse/ActiveRecord
def search(status, timeout, condition)
2018-03-17 18:26:18 +05:30
loop do
jobs = Ci::Build.where(status: status)
2018-12-05 23:21:45 +05:30
.where(condition, timeout.ago)
2020-03-13 15:44:24 +05:30
.includes(:tags, :runner, project: [:namespace, :route])
2018-03-17 18:26:18 +05:30
.limit(100)
.to_a
break if jobs.empty?
jobs.each do |job|
2020-03-13 15:44:24 +05:30
with_context(project: job.project) { yield(job) }
2018-03-17 18:26:18 +05:30
end
2017-08-17 22:00:37 +05:30
end
end
2018-12-05 23:21:45 +05:30
# rubocop: enable CodeReuse/ActiveRecord
2017-08-17 22:00:37 +05:30
2018-12-05 23:21:45 +05:30
def drop_build(type, build, status, timeout, reason)
2020-11-24 15:15:51 +05:30
Gitlab::AppLogger.info "#{self.class}: Dropping #{type} build #{build.id} for runner #{build.runner_id} (status: #{status}, timeout: #{timeout}, reason: #{reason})"
2017-08-17 22:00:37 +05:30
Gitlab::OptimisticLocking.retry_lock(build, 3) do |b|
2018-12-05 23:21:45 +05:30
b.drop(reason)
2017-08-17 22:00:37 +05:30
end
2019-12-26 22:10:19 +05:30
rescue => ex
build.doom!
track_exception_for_build(ex, build)
end
def track_exception_for_build(ex, build)
2020-01-01 13:55:28 +05:30
Gitlab::ErrorTracking.track_exception(ex,
2019-12-26 22:10:19 +05:30
build_id: build.id,
build_name: build.name,
build_stage: build.stage,
pipeline_id: build.pipeline_id,
project_id: build.project_id
2020-01-01 13:55:28 +05:30
)
2017-08-17 22:00:37 +05:30
end
end