debian-mirror-gitlab/lib/gitlab/database/migrations/background_migration_helpers.rb

196 lines
8.8 KiB
Ruby
Raw Normal View History

2020-07-28 23:09:34 +05:30
# frozen_string_literal: true
module Gitlab
module Database
module Migrations
module BackgroundMigrationHelpers
2021-04-17 20:07:23 +05:30
BATCH_SIZE = 1_000 # Number of rows to process per job
JOB_BUFFER_SIZE = 1_000 # Number of jobs to bulk queue at a time
2020-07-28 23:09:34 +05:30
2021-02-22 17:27:13 +05:30
# Queues background migration jobs for an entire table in batches.
# The default batching column used is the standard primary key `id`.
2020-07-28 23:09:34 +05:30
# Each job is scheduled with a `delay_interval` in between.
# If you use a small interval, then some jobs may run at the same time.
#
# model_class - The table or relation being iterated over
# job_class_name - The background migration job class as a string
# delay_interval - The duration between each job's scheduled time (must respond to `to_f`)
# batch_size - The maximum number of rows per job
# other_arguments - Other arguments to send to the job
# track_jobs - When this flag is set, creates a record in the background_migration_jobs table for each job that
# is scheduled to be run. These records can be used to trace execution of the background job, but there is no
# builtin support to manage that automatically at this time. You should only set this flag if you are aware of
# how it works, and intend to manually cleanup the database records in your background job.
2021-02-22 17:27:13 +05:30
# primary_column_name - The name of the primary key column if the primary key is not `id`
2020-07-28 23:09:34 +05:30
#
# *Returns the final migration delay*
#
# Example:
#
# class Route < ActiveRecord::Base
# include EachBatch
# self.table_name = 'routes'
# end
#
# queue_background_migration_jobs_by_range_at_intervals(Route, 'ProcessRoutes', 1.minute)
#
# Where the model_class includes EachBatch, and the background migration exists:
#
# class Gitlab::BackgroundMigration::ProcessRoutes
# def perform(start_id, end_id)
# # do something
# end
# end
2021-04-17 20:07:23 +05:30
def queue_background_migration_jobs_by_range_at_intervals(model_class, job_class_name, delay_interval, batch_size: BATCH_SIZE, other_job_arguments: [], initial_delay: 0, track_jobs: false, primary_column_name: :id)
2021-02-22 17:27:13 +05:30
raise "#{model_class} does not have an ID column of #{primary_column_name} to use for batch ranges" unless model_class.column_names.include?(primary_column_name.to_s)
raise "#{primary_column_name} is not an integer column" unless model_class.columns_hash[primary_column_name.to_s].type == :integer
2020-07-28 23:09:34 +05:30
2022-03-02 08:16:31 +05:30
job_coordinator = coordinator_for_tracking_database
2020-07-28 23:09:34 +05:30
# To not overload the worker too much we enforce a minimum interval both
# when scheduling and performing jobs.
2022-03-02 08:16:31 +05:30
delay_interval = [delay_interval, job_coordinator.minimum_interval].max
2020-07-28 23:09:34 +05:30
final_delay = 0
2021-03-08 18:12:59 +05:30
batch_counter = 0
2020-07-28 23:09:34 +05:30
2021-11-18 22:05:49 +05:30
model_class.each_batch(of: batch_size, column: primary_column_name) do |relation, index|
2021-09-30 23:02:18 +05:30
max = relation.arel_table[primary_column_name].maximum
min = relation.arel_table[primary_column_name].minimum
start_id, end_id = relation.pluck(min, max).first
2020-07-28 23:09:34 +05:30
2022-03-02 08:16:31 +05:30
# `SingleDatabaseWorker.bulk_perform_in` schedules all jobs for
2020-07-28 23:09:34 +05:30
# the same time, which is not helpful in most cases where we wish to
# spread the work over time.
final_delay = initial_delay + delay_interval * index
full_job_arguments = [start_id, end_id] + other_job_arguments
track_in_database(job_class_name, full_job_arguments) if track_jobs
2022-03-02 08:16:31 +05:30
migrate_in(final_delay, job_class_name, full_job_arguments, coordinator: job_coordinator)
2021-03-08 18:12:59 +05:30
batch_counter += 1
2020-07-28 23:09:34 +05:30
end
2021-03-08 18:12:59 +05:30
duration = initial_delay + delay_interval * batch_counter
say <<~SAY
Scheduled #{batch_counter} #{job_class_name} jobs with a maximum of #{batch_size} records per batch and an interval of #{delay_interval} seconds.
The migration is expected to take at least #{duration} seconds. Expect all jobs to have completed after #{Time.zone.now + duration}."
SAY
2020-07-28 23:09:34 +05:30
final_delay
end
2021-09-04 01:27:46 +05:30
# Requeue pending jobs previously queued with #queue_background_migration_jobs_by_range_at_intervals
#
# This method is useful to schedule jobs that had previously failed.
2022-01-26 12:08:38 +05:30
# It can only be used if the previous background migration used job tracking like the queue_background_migration_jobs_by_range_at_intervals helper.
2021-09-04 01:27:46 +05:30
#
# job_class_name - The background migration job class as a string
# delay_interval - The duration between each job's scheduled time
# batch_size - The maximum number of jobs to fetch to memory from the database.
def requeue_background_migration_jobs_by_range_at_intervals(job_class_name, delay_interval, batch_size: BATCH_SIZE, initial_delay: 0)
2022-03-02 08:16:31 +05:30
job_coordinator = coordinator_for_tracking_database
2021-09-04 01:27:46 +05:30
# To not overload the worker too much we enforce a minimum interval both
# when scheduling and performing jobs.
2022-03-02 08:16:31 +05:30
delay_interval = [delay_interval, job_coordinator.minimum_interval].max
2021-09-04 01:27:46 +05:30
final_delay = 0
job_counter = 0
jobs = Gitlab::Database::BackgroundMigrationJob.pending.where(class_name: job_class_name)
jobs.each_batch(of: batch_size) do |job_batch|
job_batch.each do |job|
final_delay = initial_delay + delay_interval * job_counter
2022-03-02 08:16:31 +05:30
migrate_in(final_delay, job_class_name, job.arguments, coordinator: job_coordinator)
2021-09-04 01:27:46 +05:30
job_counter += 1
end
end
duration = initial_delay + delay_interval * job_counter
say <<~SAY
Scheduled #{job_counter} #{job_class_name} jobs with an interval of #{delay_interval} seconds.
The migration is expected to take at least #{duration} seconds. Expect all jobs to have completed after #{Time.zone.now + duration}."
SAY
duration
end
2021-10-27 15:23:28 +05:30
# Force a background migration to complete.
#
# WARNING: This method will block the caller and move the background migration from an
# asynchronous migration to a synchronous migration.
#
# 1. Steal work from sidekiq and perform immediately (avoid duplicates generated by step 2).
# 2. Process any pending tracked jobs.
# 3. Steal work from sidekiq and perform immediately (clear anything left from step 2).
# 4. Optionally remove job tracking information.
#
# This method does not garauntee that all jobs completed successfully.
2022-01-26 12:08:38 +05:30
# It can only be used if the previous background migration used the queue_background_migration_jobs_by_range_at_intervals helper.
2021-10-27 15:23:28 +05:30
def finalize_background_migration(class_name, delete_tracking_jobs: ['succeeded'])
2022-03-02 08:16:31 +05:30
job_coordinator = coordinator_for_tracking_database
2021-10-27 15:23:28 +05:30
# Empty the sidekiq queue.
2022-03-02 08:16:31 +05:30
job_coordinator.steal(class_name)
2021-10-27 15:23:28 +05:30
# Process pending tracked jobs.
jobs = Gitlab::Database::BackgroundMigrationJob.pending.for_migration_class(class_name)
2022-03-02 08:16:31 +05:30
2021-10-27 15:23:28 +05:30
jobs.find_each do |job|
2022-03-02 08:16:31 +05:30
job_coordinator.perform(job.class_name, job.arguments)
2021-10-27 15:23:28 +05:30
end
# Empty the sidekiq queue.
2022-03-02 08:16:31 +05:30
job_coordinator.steal(class_name)
2021-10-27 15:23:28 +05:30
# Delete job tracking rows.
delete_job_tracking(class_name, status: delete_tracking_jobs) if delete_tracking_jobs
end
2022-03-02 08:16:31 +05:30
def migrate_in(*args, coordinator: coordinator_for_tracking_database)
2020-07-28 23:09:34 +05:30
with_migration_context do
2022-03-02 08:16:31 +05:30
coordinator.perform_in(*args)
2020-07-28 23:09:34 +05:30
end
end
2021-04-29 21:17:54 +05:30
def delete_queued_jobs(class_name)
2022-03-02 08:16:31 +05:30
coordinator_for_tracking_database.steal(class_name) do |job|
2021-04-29 21:17:54 +05:30
job.delete
false
end
end
2021-10-27 15:23:28 +05:30
def delete_job_tracking(class_name, status: 'succeeded')
status = Array(status).map { |s| Gitlab::Database::BackgroundMigrationJob.statuses[s] }
jobs = Gitlab::Database::BackgroundMigrationJob.where(status: status).for_migration_class(class_name)
jobs.each_batch { |batch| batch.delete_all }
end
2020-07-28 23:09:34 +05:30
private
2022-03-02 08:16:31 +05:30
def with_migration_context(&block)
Gitlab::ApplicationContext.with_context(caller_id: self.class.to_s, &block)
end
2020-07-28 23:09:34 +05:30
def track_in_database(class_name, arguments)
Gitlab::Database::BackgroundMigrationJob.create!(class_name: class_name, arguments: arguments)
end
2022-03-02 08:16:31 +05:30
def coordinator_for_tracking_database
Gitlab::BackgroundMigration.coordinator_for_database(tracking_database)
end
def tracking_database
Gitlab::BackgroundMigration::DEFAULT_TRACKING_DATABASE
end
2020-07-28 23:09:34 +05:30
end
end
end
end