gitlab-org--gitlab-foss/doc/development/uploads.md

27 KiB

stage group info
none unassigned To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments

Uploads development documentation

GitLab Workhorse has special rules for handling uploads. We process the upload in Workhorse to prevent occupying a Ruby process on I/O operations and because it is cheaper. This process can also directly upload to object storage.

The problem description

The following graph explains machine boundaries in a scalable GitLab installation. Without any Workhorse optimization in place, we can expect incoming requests to follow the numbers on the arrows.

graph TB
    subgraph "load balancers"
      LB(Proxy)
    end

    subgraph "Shared storage"
       nfs(NFS)
    end

    subgraph "redis cluster"
       r(persisted redis)
    end
    LB-- 1 -->Workhorse

    subgraph "web or API fleet"
      Workhorse-- 2 -->rails
    end
    rails-- "3 (write files)" -->nfs
    rails-- "4 (schedule a job)" -->r

    subgraph sidekiq
      s(sidekiq)
    end
    s-- "5 (fetch a job)" -->r
    s-- "6 (read files)" -->nfs

We have three challenges here: performance, availability, and scalability.

Performance

Rails process are expensive in terms of both CPU and memory. Ruby global interpreter lock adds to cost too because the Ruby process spends time on I/O operations on step 3 causing incoming requests to pile up.

In order to improve this, disk buffered upload was implemented. With this, Rails no longer deals with writing uploaded files to disk.

graph TB
    subgraph "load balancers"
      LB(HA Proxy)
    end

    subgraph "Shared storage"
       nfs(NFS)
    end

    subgraph "redis cluster"
       r(persisted redis)
    end
    LB-- 1 -->Workhorse

    subgraph "web or API fleet"
      Workhorse-- "3 (without files)" -->rails
    end
    Workhorse -- "2 (write files)" -->nfs
    rails-- "4 (schedule a job)" -->r

    subgraph sidekiq
      s(sidekiq)
    end
    s-- "5 (fetch a job)" -->r
    s-- "6 (read files)" -->nfs

Availability

There's also an availability problem in this setup, NFS is a single point of failure.

To address this problem an HA object storage can be used and it's supported by direct upload

Scalability

Scaling NFS is outside of our support scope, and NFS is not a part of cloud native installations.

All features that require Sidekiq and do not use direct upload doesn't work without NFS. In Kubernetes, machine boundaries translate to PODs, and in this case the uploaded file is written into the POD private disk. Since Sidekiq POD cannot reach into other pods, the operation fails to read it.

How to select the proper level of acceleration?

Selecting the proper acceleration is a tradeoff between speed of development and operational costs.

We can identify three major use-cases for an upload:

  1. storage: if we are uploading for storing a file (like artifacts, packages, or discussion attachments). In this case direct upload is the proper level as it's the less resource-intensive operation. Additional information can be found on File Storage in GitLab.
  2. in-controller/synchronous processing: if we allow processing small files synchronously, using disk buffered upload may speed up development.
  3. Sidekiq/asynchronous processing: Asynchronous processing must implement direct upload, the reason being that it's the only way to support Cloud Native deployments without a shared NFS.

For more details about currently broken feature see epic &1802.

Handling repository uploads

Some features involves Git repository uploads without using a regular Git client. Some examples are uploading a repository file from the web interface and design management.

Those uploads requires the rails controller to act as a Git client in lieu of the user. Those operation falls into in-controller/synchronous processing category, but we have no warranties on the file size.

In case of a LFS upload, the file pointer is committed synchronously, but file upload to object storage is performed asynchronously with Sidekiq.

Upload encodings

By upload encoding we mean how the file is included within the incoming request.

We have three kinds of file encoding in our uploads:

  1. multipart: multipart/form-data is the most common, a file is encoded as a part of a multipart encoded request.
  2. body: some APIs uploads files as the whole request body.
  3. JSON: some JSON APIs upload files as base64-encoded strings. This requires a change to GitLab Workhorse, which is tracked in this issue.

Uploading technologies

By uploading technologies we mean how all the involved services interact with each other.

GitLab supports 3 kinds of uploading technologies, here follows a brief description with a sequence diagram for each one. Diagrams are not meant to be exhaustive.

Rack Multipart upload

This is the default kind of upload, and it's the most expensive in terms of resources.

In this case, Workhorse is unaware of files being uploaded and acts as a regular proxy.

When a multipart request reaches the rails application, Rack::Multipart leaves behind temporary files in /tmp and uses valuable Ruby process time to copy files around.

sequenceDiagram
    participant c as Client
    participant w as Workhorse
    participant r as Rails

    activate c
    c ->>+w: POST /some/url/upload
    w->>+r:  POST /some/url/upload

    r->>r: save the incoming file on /tmp
    r->>r: read the file for processing

    r-->>-c: request result
    deactivate c
    deactivate w

Disk buffered upload

This kind of upload avoids wasting resources caused by handling upload writes to /tmp in rails.

This optimization is not active by default on REST API requests.

When enabled, Workhorse looks for files in multipart MIME requests, uploading any it finds to a temporary file on shared storage. The MIME data in the request is replaced with the path to the corresponding file before it is forwarded to Rails.

To prevent abuse of this feature, Workhorse signs the modified request with a special header, stating which entries it modified. Rails ignores any unsigned path entries.

sequenceDiagram
    participant c as Client
    participant w as Workhorse
    participant r as Rails
    participant s as NFS

    activate c
    c ->>+w: POST /some/url/upload

    w->>+s: save the incoming file on a temporary location
    s-->>-w: request result

    w->>+r:  POST /some/url/upload
    Note over w,r: file was replaced with its location<br>and other metadata

    opt requires async processing
      r->>+redis: schedule a job
      redis-->>-r: job is scheduled
    end

    r-->>-c: request result
    deactivate c
    w->>-w: cleanup

    opt requires async processing
      activate sidekiq
      sidekiq->>+redis: fetch a job
      redis-->>-sidekiq: job

      sidekiq->>+s: read file
      s-->>-sidekiq: file

      sidekiq->>sidekiq: process file

      deactivate sidekiq
    end

Direct upload

This is the more advanced acceleration technique we have in place.

Workhorse asks Rails for temporary pre-signed object storage URLs and directly uploads to object storage.

In this setup, an extra Rails route must be implemented in order to handle authorization. Examples of this can be found in:

Direct upload falls back to disk buffered upload when direct_upload is disabled inside the object storage setting. The answer to the /authorize call contains only a file system path.

sequenceDiagram
    participant c as Client
    participant w as Workhorse
    participant r as Rails
    participant os as Object Storage

    activate c
    c ->>+w: POST /some/url/upload

    w ->>+r: POST /some/url/upload/authorize
    Note over w,r: this request has an empty body
    r-->>-w: presigned OS URL

    w->>+os: PUT file
    Note over w,os: file is stored on a temporary location. Rails select the destination
    os-->>-w: request result

    w->>+r:  POST /some/url/upload
    Note over w,r: file was replaced with its location<br>and other metadata

    r->>+os: move object to final destination
    os-->>-r: request result

    opt requires async processing
      r->>+redis: schedule a job
      redis-->>-r: job is scheduled
    end

    r-->>-c: request result
    deactivate c
    w->>-w: cleanup

    opt requires async processing
      activate sidekiq
      sidekiq->>+redis: fetch a job
      redis-->>-sidekiq: job

      sidekiq->>+os: get object
      os-->>-sidekiq: file

      sidekiq->>sidekiq: process file

      deactivate sidekiq
    end

How to add a new upload route

In this section, we describe how to add a new upload route accelerated by Workhorse for body and multipart encoded uploads.

Upload routes belong to one of these categories:

  1. Rails controllers: uploads handled by Rails controllers.
  2. Grape API: uploads handled by a Grape API endpoint.
  3. GraphQL API: uploads handled by a GraphQL resolve function.

WARNING: GraphQL uploads do not support direct upload yet. Depending on the use case, the feature may not work on installations without NFS (like GitLab.com or Kubernetes installations). Uploading to object storage inside the GraphQL resolve function may result in timeout errors. For more details please follow issue #280819.

Update Workhorse for the new route

For both the Rails controller and Grape API uploads, Workhorse has to be updated in order to get the support for the new upload route.

  1. Open a new issue in the Workhorse tracker describing precisely the new upload route:
    • The route's URL.
    • The upload encoding.
    • If possible, provide a dump of the upload request.
  2. Implement and get the MR merged for this issue above.
  3. Ask the Maintainers of Workhorse to create a new release. You can do that in the MR directly during the maintainer review or ask for it in the #workhorse Slack channel.
  4. Bump the Workhorse version file to the version you have from the previous points, or bump it in the same merge request that contains the Rails changes (see Implementing the new route with a Rails controller or Implementing the new route with a Grape API endpoint below).

Implementing the new route with a Rails controller

For a Rails controller upload, we usually have a multipart upload and there are a few things to do:

  1. The upload is available under the parameter name you're using. For example, it could be an artifact or a nested parameter such as user[avatar]. Let's say that we have the upload under the file parameter, reading params[:file] should get you an UploadedFile instance.
  2. Generally speaking, it's a good idea to check if the instance is from the UploadedFile class. For example, see how we checked that the parameter is indeed an UploadedFile.

WARNING: Do not call UploadedFile#from_params directly! Do not build an UploadedFile instance using UploadedFile#from_params! This method can be unsafe to use depending on the params passed. Instead, use the UploadedFile instance that multipart.rb builds automatically for you.

Implementing the new route with a Grape API endpoint

For a Grape API upload, we can have body or a multipart upload. Things are slightly more complicated: two endpoints are needed. One for the Workhorse pre-upload authorization and one for accepting the upload metadata from Workhorse:

  1. Implement an endpoint with the URL + /authorize suffix that will:
    • Check that the request is coming from Workhorse with the require_gitlab_workhorse! from the API helpers.
    • Check user permissions.
    • Set the status to 200 with status 200.
    • Set the content type with content_type Gitlab::Workhorse::INTERNAL_API_CONTENT_TYPE.
    • Use your dedicated Uploader class (let's say that it's FileUploader) to build the response with FileUploader.workhorse_authorize(params).
  2. Implement the endpoint for the upload request that will:
    • Require all the UploadedFile objects as parameters.
      • For example, if we expect a single parameter file to be an UploadedFile instance, use requires :file, type: ::API::Validations::Types::WorkhorseFile.
      • Body upload requests have their upload available under the parameter file.
    • Check that the request is coming from Workhorse with the require_gitlab_workhorse! from the API helpers.
    • Check the user permissions.
    • The remaining code of the processing. This is where the code must be reading the parameter (for our example, it would be params[:file]).

WARNING: Do not call UploadedFile#from_params directly! Do not build an UploadedFile object using UploadedFile#from_params! This method can be unsafe to use depending on the params passed. Instead, use the UploadedFile object that multipart.rb builds automatically for you.

Document Object Storage buckets and CarrierWave integration

When using Object Storage, GitLab expects each kind of upload to maintain its own bucket in the respective Object Storage destination. Moreover, the integration with CarrierWave is not used all the time. The Object Storage Working Group is investigating an approach that unifies Object Storage buckets into a single one and removes CarrierWave so as to simplify implementation and administration of uploads.

Therefore, document new uploads here by slotting them into the following tables:

Feature bucket details

Feature Upload technology Uploader Bucket structure
Job artifacts direct upload workhorse /artifacts/<proj_id_hash>/<date>/<job_id>/<artifact_id>
Pipeline artifacts carrierwave sidekiq /artifacts/<proj_id_hash>/pipelines/<pipeline_id>/artifacts/<artifact_id>
Live job traces fog sidekiq /artifacts/tmp/builds/<job_id>/chunks/<chunk_index>.log
Job traces archive carrierwave sidekiq /artifacts/<proj_id_hash>/<date>/<job_id>/<artifact_id>/job.log
Autoscale runner caching N/A gitlab-runner /gitlab-com-[platform-]runners-cache/???
Backups N/A s3cmd, awscli, or gcs /gitlab-backups/???
Git LFS direct upload workhorse /lsf-objects/<lfs_obj_oid[0:2]>/<lfs_obj_oid[2:2]>
Design management files disk buffering rails controller /lsf-objects/<lfs_obj_oid[0:2]>/<lfs_obj_oid[2:2]>
Design management thumbnails carrierwave sidekiq /uploads/design_management/action/image_v432x230/<model_id>
Generic file uploads direct upload workhorse /uploads/@hashed/[0:2]/[2:4]/<hash1>/<hash2>/file
Generic file uploads - personal snippets direct upload workhorse /uploads/personal_snippet/<snippet_id>/<filename>
Global appearance settings disk buffering rails controller /uploads/appearance/...
Topics disk buffering rails controller /uploads/projects/topic/...
Avatar images direct upload workhorse /uploads/[user,group,project]/avatar/<model_id>
Import/export direct upload workhorse /uploads/import_export_upload/???
GitLab Migration carrierwave sidekiq /uploads/bulk_imports/???
MR diffs carrierwave sidekiq /external-diffs/merge_request_diffs/mr-<mr_id>/diff-<diff_id>
Package manager archives direct upload sidekiq /packages/<proj_id_hash>/packages/<pkg_segment>/files/<pkg_file_id>
Package manager archives direct upload sidekiq /packages/<container_id_hash>/debian_*_component_file/<component_file_id>
Package manager archives direct upload sidekiq /packages/<container_id_hash>/debian_*_distribution/<distribution_file_id>
Container image cache (?) direct upload workhorse /dependency-proxy/<group_id_hash>/dependency_proxy/<group_id>/files/<proxy_id>/<blob_id or manifest_id>
Terraform state files carrierwave rails controller /terraform/<proj_id_hash>/<terraform_state_id>
Pages content archives carrierwave sidekiq /gitlab-gprd-pages/<proj_id_hash>/pages_deployments/<deployment_id>/
Secure Files carrierwave sidekiq /ci-secure-files/<proj_id_hash>/secure_files/<secure_file_id>/

CarrierWave integration

File Carrierwave usage Categorized
app/models/project.rb include Avatarable
app/models/projects/topic.rb include Avatarable
app/models/group.rb include Avatarable
app/models/user.rb include Avatarable
app/models/terraform/state_version.rb include FileStoreMounter
app/models/ci/job_artifact.rb include FileStoreMounter
app/models/ci/pipeline_artifact.rb include FileStoreMounter
app/models/pages_deployment.rb include FileStoreMounter
app/models/lfs_object.rb include FileStoreMounter
app/models/dependency_proxy/blob.rb include FileStoreMounter
app/models/dependency_proxy/manifest.rb include FileStoreMounter
app/models/packages/composer/cache_file.rb include FileStoreMounter
app/models/packages/package_file.rb include FileStoreMounter
app/models/concerns/packages/debian/component_file.rb include FileStoreMounter
ee/app/models/issuable_metric_image.rb include FileStoreMounter
ee/app/models/vulnerabilities/remediation.rb include FileStoreMounter
ee/app/models/vulnerabilities/export.rb include FileStoreMounter
app/models/packages/debian/project_distribution.rb include Packages::Debian::Distribution
app/models/packages/debian/group_distribution.rb include Packages::Debian::Distribution
app/models/packages/debian/project_component_file.rb include Packages::Debian::ComponentFile
app/models/packages/debian/group_component_file.rb include Packages::Debian::ComponentFile
app/models/merge_request_diff.rb mount_uploader :external_diff, ExternalDiffUploader
app/models/note.rb mount_uploader :attachment, AttachmentUploader
app/models/appearance.rb mount_uploader :logo, AttachmentUploader
app/models/appearance.rb mount_uploader :header_logo, AttachmentUploader
app/models/appearance.rb mount_uploader :favicon, FaviconUploader
app/models/project.rb mount_uploader :bfg_object_map, AttachmentUploader
app/models/import_export_upload.rb mount_uploader :import_file, ImportExportUploader
app/models/import_export_upload.rb mount_uploader :export_file, ImportExportUploader
app/models/ci/deleted_object.rb mount_uploader :file, DeletedObjectUploader
app/models/design_management/action.rb mount_uploader :image_v432x230, DesignManagement::DesignV432x230Uploader
app/models/concerns/packages/debian/distribution.rb mount_uploader :signed_file, Packages::Debian::DistributionReleaseFileUploader
app/models/bulk_imports/export_upload.rb mount_uploader :export_file, ExportUploader
ee/app/models/user_permission_export_upload.rb mount_uploader :file, AttachmentUploader
app/models/ci/secure_file.rb include FileStoreMounter