Commit 2122d722 authored by Kamil Trzciński's avatar Kamil Trzciński

Merge branch 'fix/gb/fix-redundant-pipeline-stages' into 'master'

Remove redundant pipeline stages from the database

Closes #41769

See merge request gitlab-org/gitlab-ce!16580
parents 9bd044d9 394c1c65
Pipeline #17208200 passed with stages
in 46 minutes and 8 seconds
......@@ -7,6 +7,8 @@ module Ci
# stage.
#
class EnsureStageService < BaseService
EnsureStageError = Class.new(StandardError)
def execute(build)
@build = build
......@@ -22,8 +24,16 @@ module Ci
private
def ensure_stage
def ensure_stage(attempts: 2)
find_stage || create_stage
rescue ActiveRecord::RecordNotUnique
retry if (attempts -= 1) > 0
raise EnsureStageError, <<~EOS
We failed to find or create a unique pipeline stage after 2 retries.
This should never happen and is most likely the result of a bug in
the database load balancing code.
EOS
end
def find_stage
......
module Ci
class RetryBuildService < ::BaseService
CLONE_ACCESSORS = %i[pipeline project ref tag options commands name
allow_failure stage_id stage stage_idx trigger_request
allow_failure stage stage_id stage_idx trigger_request
yaml_variables when environment coverage_regex
description tag_list protected].freeze
......
class RemoveRedundantPipelineStages < ActiveRecord::Migration
include Gitlab::Database::MigrationHelpers
DOWNTIME = false
disable_ddl_transaction!
def up(attempts: 100)
remove_redundant_pipeline_stages!
remove_outdated_index!
add_unique_index!
rescue ActiveRecord::RecordNotUnique
retry if (attempts -= 1) > 0
raise StandardError, <<~EOS
Failed to add an unique index to ci_stages, despite retrying the
migration 100 times.
See https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/16580.
EOS
end
def down
remove_concurrent_index :ci_stages, [:pipeline_id, :name], unique: true
add_concurrent_index :ci_stages, [:pipeline_id, :name]
end
private
def remove_outdated_index!
return unless index_exists?(:ci_stages, [:pipeline_id, :name])
remove_concurrent_index :ci_stages, [:pipeline_id, :name]
end
def add_unique_index!
add_concurrent_index :ci_stages, [:pipeline_id, :name], unique: true
end
def remove_redundant_pipeline_stages!
disable_statement_timeout
redundant_stages_ids = <<~SQL
SELECT id FROM ci_stages WHERE (pipeline_id, name) IN (
SELECT pipeline_id, name FROM ci_stages
GROUP BY pipeline_id, name HAVING COUNT(*) > 1
)
SQL
execute <<~SQL
UPDATE ci_builds SET stage_id = NULL WHERE stage_id IN (#{redundant_stages_ids})
SQL
if Gitlab::Database.postgresql?
execute <<~SQL
DELETE FROM ci_stages WHERE id IN (#{redundant_stages_ids})
SQL
else # We can't modify a table we are selecting from on MySQL
execute <<~SQL
DELETE a FROM ci_stages AS a, ci_stages AS b
WHERE a.pipeline_id = b.pipeline_id AND a.name = b.name
AND a.id <> b.id
SQL
end
end
end
......@@ -451,7 +451,7 @@ ActiveRecord::Schema.define(version: 20180206200543) do
t.integer "lock_version"
end
add_index "ci_stages", ["pipeline_id", "name"], name: "index_ci_stages_on_pipeline_id_and_name", using: :btree
add_index "ci_stages", ["pipeline_id", "name"], name: "index_ci_stages_on_pipeline_id_and_name", unique: true, using: :btree
add_index "ci_stages", ["pipeline_id"], name: "index_ci_stages_on_pipeline_id", using: :btree
add_index "ci_stages", ["project_id"], name: "index_ci_stages_on_project_id", using: :btree
......
require 'spec_helper'
require Rails.root.join('db', 'post_migrate', '20180119121225_remove_redundant_pipeline_stages.rb')
describe RemoveRedundantPipelineStages, :migration do
let(:projects) { table(:projects) }
let(:pipelines) { table(:ci_pipelines) }
let(:stages) { table(:ci_stages) }
let(:builds) { table(:ci_builds) }
before do
projects.create!(id: 123, name: 'gitlab', path: 'gitlab-ce')
pipelines.create!(id: 234, project_id: 123, ref: 'master', sha: 'adf43c3a')
stages.create!(id: 6, project_id: 123, pipeline_id: 234, name: 'build')
stages.create!(id: 10, project_id: 123, pipeline_id: 234, name: 'build')
stages.create!(id: 21, project_id: 123, pipeline_id: 234, name: 'build')
stages.create!(id: 41, project_id: 123, pipeline_id: 234, name: 'test')
stages.create!(id: 62, project_id: 123, pipeline_id: 234, name: 'test')
stages.create!(id: 102, project_id: 123, pipeline_id: 234, name: 'deploy')
builds.create!(id: 1, commit_id: 234, project_id: 123, stage_id: 10)
builds.create!(id: 2, commit_id: 234, project_id: 123, stage_id: 21)
builds.create!(id: 3, commit_id: 234, project_id: 123, stage_id: 21)
builds.create!(id: 4, commit_id: 234, project_id: 123, stage_id: 41)
builds.create!(id: 5, commit_id: 234, project_id: 123, stage_id: 62)
builds.create!(id: 6, commit_id: 234, project_id: 123, stage_id: 102)
end
it 'removes ambiguous stages and preserves builds' do
expect(stages.all.count).to eq 6
expect(builds.all.count).to eq 6
migrate!
expect(stages.all.count).to eq 1
expect(builds.all.count).to eq 6
expect(builds.all.pluck(:stage_id).compact).to eq [102]
end
it 'retries when incorrectly added index exception is caught' do
allow_any_instance_of(described_class)
.to receive(:remove_redundant_pipeline_stages!)
expect_any_instance_of(described_class)
.to receive(:remove_outdated_index!)
.exactly(100).times.and_call_original
expect { migrate! }
.to raise_error StandardError, /Failed to add an unique index/
end
it 'does not retry when unknown exception is being raised' do
allow(subject).to receive(:remove_outdated_index!)
expect(subject).to receive(:remove_redundant_pipeline_stages!).once
allow(subject).to receive(:add_unique_index!).and_raise(StandardError)
expect { subject.up(attempts: 3) }.to raise_error StandardError
end
end
require 'spec_helper'
describe Ci::EnsureStageService, '#execute' do
set(:project) { create(:project) }
set(:user) { create(:user) }
let(:stage) { create(:ci_stage_entity) }
let(:job) { build(:ci_build) }
let(:service) { described_class.new(project, user) }
context 'when build has a stage assigned' do
it 'does not create a new stage' do
job.assign_attributes(stage_id: stage.id)
expect { service.execute(job) }.not_to change { Ci::Stage.count }
end
end
context 'when build does not have a stage assigned' do
it 'creates a new stage' do
job.assign_attributes(stage_id: nil, stage: 'test')
expect { service.execute(job) }.to change { Ci::Stage.count }.by(1)
end
end
context 'when build is invalid' do
it 'does not create a new stage' do
job.assign_attributes(stage_id: nil, ref: nil)
expect { service.execute(job) }.not_to change { Ci::Stage.count }
end
end
context 'when new stage can not be created because of an exception' do
before do
allow(Ci::Stage).to receive(:create!)
.and_raise(ActiveRecord::RecordNotUnique.new('Duplicates!'))
end
it 'retries up to two times' do
job.assign_attributes(stage_id: nil)
expect(service).to receive(:find_stage).exactly(2).times
expect { service.execute(job) }
.to raise_error(Ci::EnsureStageService::EnsureStageError)
end
end
end
......@@ -5,7 +5,11 @@ describe Ci::RetryBuildService do
set(:project) { create(:project) }
set(:pipeline) { create(:ci_pipeline, project: project) }
let(:build) { create(:ci_build, pipeline: pipeline) }
let(:stage) do
Ci::Stage.create!(project: project, pipeline: pipeline, name: 'test')
end
let(:build) { create(:ci_build, pipeline: pipeline, stage_id: stage.id) }
let(:service) do
described_class.new(project, user)
......@@ -27,29 +31,27 @@ describe Ci::RetryBuildService do
user_id auto_canceled_by_id retried failure_reason].freeze
shared_examples 'build duplication' do
let(:stage) do
# TODO, we still do not have factory for new stages, we will need to
# switch existing factory to persist stages, instead of using LegacyStage
#
Ci::Stage.create!(project: project, pipeline: pipeline, name: 'test')
end
let(:another_pipeline) { create(:ci_empty_pipeline, project: project) }
let(:build) do
create(:ci_build, :failed, :artifacts, :expired, :erased,
:queued, :coverage, :tags, :allowed_to_fail, :on_tag,
:triggered, :trace_artifact, :teardown_environment,
description: 'my-job', stage: 'test', pipeline: pipeline,
auto_canceled_by: create(:ci_empty_pipeline, project: project)) do |build|
##
# TODO, workaround for FactoryBot limitation when having both
# stage (text) and stage_id (integer) columns in the table.
build.stage_id = stage.id
end
description: 'my-job', stage: 'test', stage_id: stage.id,
pipeline: pipeline, auto_canceled_by: another_pipeline)
end
before do
# Make sure that build has both `stage_id` and `stage` because FactoryBot
# can reset one of the fields when assigning another. We plan to deprecate
# and remove legacy `stage` column in the future.
build.update_attributes(stage: 'test', stage_id: stage.id)
end
describe 'clone accessors' do
CLONE_ACCESSORS.each do |attribute|
it "clones #{attribute} build attribute" do
expect(build.send(attribute)).not_to be_nil
expect(new_build.send(attribute)).not_to be_nil
expect(new_build.send(attribute)).to eq build.send(attribute)
end
......@@ -122,10 +124,12 @@ describe Ci::RetryBuildService do
context 'when there are subsequent builds that are skipped' do
let!(:subsequent_build) do
create(:ci_build, :skipped, stage_idx: 1, pipeline: pipeline)
create(:ci_build, :skipped, stage_idx: 2,
pipeline: pipeline,
stage: 'deploy')
end
it 'resumes pipeline processing in subsequent stages' do
it 'resumes pipeline processing in a subsequent stage' do
service.execute(build)
expect(subsequent_build.reload).to be_created
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment