Skip to content

Latest commit

 

History

History
216 lines (170 loc) · 11.5 KB

README.md

File metadata and controls

216 lines (170 loc) · 11.5 KB
Raw

Build

galc-api

An API for the UC Berkeley Library Graphic Arts Loan Collection.

For the GALC user interface, see the galc-ui repository.

Architecture

galc-api is by and large a conventional database-backed Rails API-only application. Areas in which it deviates from typical Rails practice include authentication / authorization (which uses JSON Web Tokens rather than session cookies), JSON serialization (which follows the JSON:API specification, rather than using Rails' default JSON serialization), and the handling of image files (which are managed on the filesystem rather than via ActiveStorage or similar mechanisms). Details on each of these areas are provided below.

RESTful Endpoints

Persistent resources

The application provides RESTful endpoints for the following resources, represented by conventional ActiveRecord database models:

Endpoint Model Description
/items Item Items (i.e. prints) in the GALC collection
/images Image Images of prints for display in search results
/facets Facet Search facets (read-only)
/terms Term Controlled vocabulary search terms (read-only)
/closures Closure Periods when the GALC collection is closed to checkouts

Note that the /items endpoint accepts both faceted search terms and arbitrary keywords as filter parameters following the JSON:API Filtering Recommentation, and also accepts pagination paramters following the JSON:API pagination spec.

Note also that for POSTing new images to the the /images endpoint, we use plain multipart/form-data rather than JSON.

Synthetic resources

In addition, the following endpoints appear as RESTful resources, but are not represented in the database:

Endpoint Description
/reservations Requests to reserve a GALC print (write-only)
/users Application users (current user only)
/build_info Build information (read only)
Reservations

Clients can POST a request to create a reservation, which returns the created reservation as a JSON object. However, reservations are not stored in the databse; "creating" a reservation simply sends an email to the configured reservation email address (set via the $RESERVE_EMAIL_TO environment variable).

Users

Cross-site cookie issues prevent using a conventional Rails session to manage login state. Instead, authentication/authorization information is sent with each request via an encrypted JSON Web Token in the HTTP Authorization header. If this token is present, clients can GET the current user either by UID or by the special value current — decrypting the token and returning the user information as a JSON object. (Note that the UID is only accepted if it matches the ID encoded in the token.)

Build information

Build information (BuildInfo) is captured in environment variables during the Docker image build process, and returned by this endpoint in JSON format.

Non-RESTful endpoints

In addition, the following non-RESTful endpoints are implemented by AuthController:

Endpoint Description
/auth/calnet/callback Callback URL for CalNet login
/logout Target URL for redirecting to CalNet logout
/ A static webpage providing the galc-ui interface for testing; enabled only when $SERVE_TEST_UI is set

Authentication and authorization

As noted above, authentication is somewhat tricky given that it runs on a different host from the GALC UI and browser makers have in the past few years cracked down hard on cross-site cookies. This makes Rails' traditional cookie-based session mechanism infeasible and requires us to rely on JSON Web Tokens.

Furthermore, CAS/CalNet has no provision for API-based authentication in the background; the only way to log into CalNet is to navigate to the CalNet login page.

The resulting authentication / authorization flow, then, is as follows:

  1. User POSTs a form from the UI to the API's /auth/calnet endpoint (login in routes.rb. The form includes a hidden origin field with the full UI URL to which the user should be redirected after login.
  2. The API /auth/calnet endpoint constructs a CalNet callback URL, consisting of the URL for the API /auth/calnet/callback endpoint plus a url query parameter capturing the origin value from the POST payload, and sends a 302 Found response redirecting the browser to the CalNet login page, with the API callback URL (including the captured UI callback URL) as the value of a service query parameter.
  3. The user logs into CalNet, Duo, etc. and is redirected to the callback URL provided in the service parameter - the /auth/calnet/callback endpoint.
  4. The /auth/calnet/callback endpoint reads the user's identity and other relevant fields from the Omniauth / CalNet response, constructs a User object, and serializes it as an encrypted JWT, then redirects the browser back to the UI, with the JWT appended to the UI callback URL as the token query parameter.
  5. Subsequent API calls made by the UI include the still-encrypted JWT in an Authorization header.

The UI can access relevant user information — name, email, privileges, etc. — by querying the /users endpoint while including that token, as mentioned above, but cannot itself decrypt / encrypt tokens.

Within the API application, the user information serialized and encrypted in the token is treated similarly to user information serialized in a traditional session cookie — certain endpoints require authentication, others require admin privileges, and so on.

Note: A key difference from the Library's monolithic Rails applications such as Framework, UC BEARS, and AV Player, is that unauthenticated API requests to endpoints that require authentication cannot simply be redirected to login. Instead they return 401 Unauthorized, and it is the UI's responsibility to handle the situtation appropriately. In most other respects, GALC API authorization code is similar to those other apps.

JSON serialization

GALC API endpoints follow the JSON:API specification for object serialization, using the jsonapi gem. Key differences for the developer from traditional Rails JSON serialization include:

  • rendering the :jsonapi format rather than :json
  • using the application/vnd.api+json media type rather than application/json
  • explicit serialization for all model objects rather than using the default ActiveModel serializer (see the app/serializers directory)
  • side-channel information via the meta element in the response object
  • query parameters following JSON:API's paging specification and filtering recommendation instead of following Rails / ActionController norms

Testing JSON:API responses (and, in particular, getting useful information out of test failures) can be fairly involved. Some custom RSpec matchers are included in the spec/support/matchers directory, see e.g. the ItemsController request spec for an example of their usage. They're not perfect (note for instance that the specs usually need to delete the links field from the response object before comparing) but they do help. Note that these matchers depend on the serializers in app/serializers to be working properly, so those need their own lower-level unit tests.

Image file handling

In order to simplify migration from the old GALC servlet and make it easier to troubleshoot any imaging issues, uploaded images and their generated thumbnails are managed as a plain directory (in staging and production, a named NetApp volume also accessible from vm135 as /netapp/galc_staging / /netapp/galc_production/), rather than via ActiveStorage or similar.

The Image class is a plain ActiveRecord model with a corresponding database table, but also includes code to manage its associated files — reading a Rails UploadedFile object from a form POST, writing the file to the appropriate location on disk, creating and writing a thumbnail, etc.; as well as deleting both thumbnail and full-size file when the ActiveRecord model object is destroyed.

Besides API endpoints, the ImagesController is also responsible for streaming the images themselves to the browser — not ideal for a high-volume application, but good enough for now and obviates the need for a separate image server.

Build, packaging, and deployment

The GALC API application is built via GitHub actions. The workflow in build.yml runs the following jobs on each push or pull request:

  1. build: builds a Docker image and pushes it to GitHub's ghcr.io repository, tagged with the short commit hash
  2. test: if build succeeds, pulls the image and runs tests, style checks, etc.
  3. push: if test succeeds, retags the image from (1) with the branch name (e.g. main)

For each GitHub release, the workflow in release.yml uses the short commit hash associated with the release to find and pull the docker image from step (1) above, and retags it with:

  • a semantic version number taken from the release name (actually, major, minor, and patch versions — e.g. release 6.5.4 would be tagged note only as 6.5.4 but also as 6 and 6.5) and (b) the tag latest.

Deployment

Deployment is triggered manually via the ops/docker-swarm Jenkins build. The staging and production Docker stacks are configured to pull the main and latest images, respectively.