Refuse to perform an LFS clean on projects that are fork roots

When a project is at the root of a fork network, an intersection of
technical debt means that all child forks have a claim on the LFS
objects owned by the fork parent. This makes it dangerous to remove
any LFS references in the fork root without checking all the repos
for all the children, which quickly mounts up so as to be unmanageable.

For now, just skip projects like this.

If a fork is an internal or leaf node in the fork graph, they may have
some LFS objects of their own. Those references are safe to prune.

changelogs/unreleased/244848-skip-cleanup-for-forks.yml

+---
+title: Refuse to perform an LFS clean on projects that are fork roots
+merge_request: 41703
+author:
+type: fixed

doc/raketasks/cleanup.md

@@ -10,6 +10,10 @@ DANGER: **Danger:**
 Do not run this within 12 hours of a GitLab upgrade. This is to ensure that all background migrations
 have finished, which otherwise may lead to data loss.
 
+CAUTION: **WARNING:**
+Removing LFS files from a project with forks is currently unsafe. The rake task
+will refuse to run on projects with forks.
+
 When you remove LFS files from a repository's history, they become orphaned and continue to consume
 disk space. With this Rake task, you can remove invalid references from the database, which
 will allow garbage collection of LFS files.

lib/gitlab/cleanup/orphan_lfs_file_references.rb

@@ -17,6 +17,14 @@ module Gitlab
       end
 
       def run!
+        # If this project is an LFS storage project (e.g. is the root of a fork
+        # network), what it is safe to remove depends on the sum of its forks.
+        # For now, skip cleaning up LFS for this complicated case
+        if project.forks_count > 0 && project.lfs_storage_project == project
+          log_info("Skipping orphan LFS check for #{project.name_with_namespace} as it is a fork root")
+          return
+        end
+
         log_info("Looking for orphan LFS files for project #{project.name_with_namespace}")
 
         remove_orphan_references

spec/lib/gitlab/cleanup/orphan_lfs_file_references_spec.rb

@@ -3,6 +3,8 @@
 require 'spec_helper'
 
 RSpec.describe Gitlab::Cleanup::OrphanLfsFileReferences do
+  include ProjectForksHelper
+
   let(:null_logger) { Logger.new('/dev/null') }
   let(:project) { create(:project, :repository, lfs_enabled: true) }
   let(:lfs_object) { create(:lfs_object) }
@@ -85,4 +87,42 @@ RSpec.describe Gitlab::Cleanup::OrphanLfsFileReferences do
       .to receive(:get_all_lfs_pointers)
       .and_return(oids.map { |oid| OpenStruct.new(lfs_oid: oid) })
   end
+
+  context 'LFS for forked projects' do
+    let!(:fork_root) { create(:project, :repository, lfs_enabled: true) }
+    let!(:fork_internal) { fork_project(fork_root, nil, repository: true) }
+    let!(:fork_leaf) { fork_project(fork_internal, nil, repository: true) }
+
+    let(:dry_run) { true }
+
+    context 'root node' do
+      let(:project) { fork_root }
+
+      it 'skips cleanup' do
+        expect(service).not_to receive(:remove_orphan_references)
+
+        service.run!
+      end
+    end
+
+    context 'internal node' do
+      let(:project) { fork_internal }
+
+      it 'runs cleanup' do
+        expect(service).to receive(:remove_orphan_references)
+
+        service.run!
+      end
+    end
+
+    context 'leaf node' do
+      let(:project) { fork_leaf }
+
+      it 'runs cleanup' do
+        expect(service).to receive(:remove_orphan_references)
+
+        service.run!
+      end
+    end
+  end
 end