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

55 lines
1.8 KiB
Ruby

# frozen_string_literal: true
class GitlabServicePingWorker # rubocop:disable Scalability/IdempotentWorker
LEASE_KEY = 'gitlab_service_ping_worker:ping'
LEASE_TIMEOUT = 86400
include ApplicationWorker
data_consistency :always
include CronjobQueue # rubocop:disable Scalability/CronWorkerContext
include Gitlab::ExclusiveLeaseHelpers
feature_category :service_ping
worker_resource_boundary :cpu
sidekiq_options retry: 3, dead: false
sidekiq_retry_in { |count| (count + 1) * 8.hours.to_i }
def perform(options = {})
# Sidekiq does not support keyword arguments, so the args need to be
# passed the old pre-Ruby 2.0 way.
#
# See https://github.com/mperham/sidekiq/issues/2372
triggered_from_cron = options.fetch('triggered_from_cron', true)
skip_db_write = options.fetch('skip_db_write', false)
# Disable service ping for GitLab.com unless called manually
# See https://gitlab.com/gitlab-org/gitlab/-/issues/292929 for details
return if Gitlab.com? && triggered_from_cron
# Multiple Sidekiq workers could run this. We should only do this at most once a day.
in_lock(LEASE_KEY, ttl: LEASE_TIMEOUT) do
# Splay the request over a minute to avoid thundering herd problems.
sleep(rand(0.0..60.0).round(3))
ServicePing::SubmitService.new(payload: usage_data, skip_db_write: skip_db_write).execute
end
end
def usage_data
ServicePing::BuildPayload.new.execute.tap do |payload|
record = {
recorded_at: payload[:recorded_at],
payload: payload,
created_at: Time.current,
updated_at: Time.current
}
RawUsageData.upsert(record, unique_by: :recorded_at)
end
rescue StandardError => err
Gitlab::ErrorTracking.track_and_raise_for_dev_exception(err)
nil
end
end