• Disk storage
  • Path segments
• Object Storage
  • Using ObjectStorage::Extension::RecordsUploads
  • Using a mounted uploader

File Storage in GitLab

We use the CarrierWave gem to handle file upload, store and retrieval.

File uploads should be accelerated by workhorse, for details refer to uploads development documentation.

There are many places where file uploading is used, according to contexts:

• System
  • Instance Logo (logo visible in sign in/sign up pages)
  • Header Logo (one displayed in the navigation bar)
• Group
  • Group avatars
• User
  • User avatars
  • User snippet attachments
• Project
  • Project avatars
  • Issues/MR/Notes Markdown attachments
  • Issues/MR/Notes Legacy Markdown attachments
  • CI Artifacts (archive, metadata, trace)
  • LFS Objects
  • Merge request diffs
  • Design Management design thumbnails
• Topic
  • Topic avatars

Disk storage

GitLab started saving everything on local disk. While directory location changed from previous versions, they are still not 100% standardized. You can see them below:

CI Artifacts and LFS Objects behave differently in CE and EE. In CE they inherit the GitlabUploader while in EE they inherit the ObjectStorage and store files in and S3 API compatible object store.

Attachments for issues, merge requests (MR), and notes in Markdown use hashed storage with the hash of the project ID.

We provide an all-in-one Rake task to migrate all uploads to object storage in one go. If a new Uploader class or model type is introduced, make sure you add a Rake task invocation corresponding to it to the category list.

Path segments

Files are stored at multiple locations and use different path schemes. All the GitlabUploader derived classes should comply with this path segment schema:

|   GitlabUploader
| ----------------------- + ------------------------- + --------------------------------- + -------------------------------- |
| `<gitlab_root>/public/` | `uploads/-/system/`       | `user/avatar/:id/`                | `:filename`                      |
| ----------------------- + ------------------------- + --------------------------------- + -------------------------------- |
| `CarrierWave.root`      | `GitlabUploader.base_dir` | `GitlabUploader#dynamic_segment`  | `CarrierWave::Uploader#filename` |
|                         | `CarrierWave::Uploader#store_dir`                             |                                  |

|   FileUploader
| ----------------------- + ------------------------- + --------------------------------- + -------------------------------- |
| `<gitlab_root>/shared/` | `artifacts/`              | `:year_:month/:id`                | `:filename`                      |
| `<gitlab_root>/shared/` | `snippets/`               | `:secret/`                        | `:filename`                      |
| ----------------------- + ------------------------- + --------------------------------- + -------------------------------- |
| `CarrierWave.root`      | `GitlabUploader.base_dir` | `GitlabUploader#dynamic_segment`  | `CarrierWave::Uploader#filename` |
|                         | `CarrierWave::Uploader#store_dir`                             |                                  |
|                         |                           | `FileUploader#upload_path`                                           |

|   ObjectStore::Concern (store = remote)
| ----------------------- + ------------------------- + ----------------------------------- + -------------------------------- |
| `<bucket_name>`         | <ignored>                 | `user/avatar/:id/`                  | `:filename`                      |
| ----------------------- + ------------------------- + ----------------------------------- + -------------------------------- |
| `#fog_dir`              | `GitlabUploader.base_dir` | `GitlabUploader#dynamic_segment`    | `CarrierWave::Uploader#filename` |
|                         |                           | `ObjectStorage::Concern#store_dir`  |                                  |
|                         |                           | `ObjectStorage::Concern#upload_path`                                   |

The RecordsUploads::Concern concern creates an Upload entry for every file stored by a GitlabUploader persisting the dynamic parts of the path using GitlabUploader#dynamic_path. You may then use the Upload#build_uploader method to manipulate the file.

Object Storage

By including the ObjectStorage::Concern in the GitlabUploader derived class, you may enable the object storage for this uploader. To enable the object storage in your uploader, you need to either 1) include RecordsUpload::Concern and prepend ObjectStorage::Extension::RecordsUploads or 2) mount the uploader and create a new field named <mount>_store.

The CarrierWave::Uploader#store_dir is overridden to

• GitlabUploader.base_dir + GitlabUploader.dynamic_segment when the store is LOCAL
• GitlabUploader.dynamic_segment when the store is REMOTE (the bucket name is used to namespace)

Using ObjectStorage::Extension::RecordsUploads

This concern includes RecordsUploads::Concern if not already included.

The ObjectStorage::Concern uploader searches for the matching Upload to select the correct object store. The Upload is mapped using #store_dirs + identifier for each store (LOCAL/REMOTE).

class SongUploader < GitlabUploader
  include RecordsUploads::Concern
  include ObjectStorage::Concern
  prepend ObjectStorage::Extension::RecordsUploads

  ...
end

class Thing < ActiveRecord::Base
  mount :theme, SongUploader # we have a great theme song!

  ...
end

Using a mounted uploader

The ObjectStorage::Concern queries the model.<mount>_store attribute to select the correct object store. This column must be present in the model schema.

class SongUploader < GitlabUploader
  include ObjectStorage::Concern

  ...
end

class Thing < ActiveRecord::Base
  attr_reader :theme_store # this is an ActiveRecord attribute
  mount :theme, SongUploader # we have a great theme song!

  def theme_store
    super || ObjectStorage::Store::LOCAL
  end

  ...
end