Writing a store

wardn ships stores for SQL databases and MongoDB. For anything else (DynamoDB, Spanner, an internal service), implement wardn.Store. The interface is small because stores only save and load records; all the authorization logic, including security, stays in wardn.

The interface

type Store interface {
  AssignmentStore // Assign, Revoke, Assignments, Members, RoleInUse, RevokeRole,
                  // ExpiredUnreported, MarkExpiryReported, RevokeExpired
  RoleStore       // CreateRole, UpdateRole, DeleteRole, Roles, RolesOwnedBy
  ScopeStore      // Link, Parent
}

// Optional: stores with a schema. wardn.New calls Migrate.
type Migrator interface {
  Migrate(ctx context.Context) error
}

The method documentation (go doc github.com/MogRex-Limited/wardn Store) states each contract precisely.

The rules that matter most

• Compare IDs exactly, byte for byte. Never case-insensitively, never normalised, never trimmed.
• Return only what was asked for. Assignments must never return another user’s or another scope’s rows. (wardn re-checks anyway, but your store must be correct.)
• Make two things atomic: creating a role whose (Owner, NameKey) is taken must fail with ErrRoleNameTaken, and UpdateRole must succeed only if the stored version equals the expected one (ErrConflict otherwise). A unique index and a conditional update do both.
• Never widen a time window. If you round times, round starts later and ends earlier.
• Honour context cancellation, and fail completely or not at all.
• RevokeExpired and MarkExpiryReported must ignore an assignment whose window changed, for example because it was assigned again.

Prove it

import "github.com/MogRex-Limited/wardn/storetest"

func TestMyStore(t *testing.T) {
  storetest.Run(t, func(t *testing.T) wardn.Store {
    return mystore.New(freshDatabase(t)) // a new, empty store per test
  })
}

The suite runs the same 25 checks the official stores pass on every change. Among them: exact identity across case, Unicode forms and widths; IDs containing SQL injection text; atomic name uniqueness and versioned updates under 8 concurrent writers; time round-trips; cancellation; and an end-to-end scenario through a real Warden.

storetest.Bench measures your store on a bank-sized dataset, with the same benchmarks as the official stores (Performance).

Note

A shared cache works the same way: implement wardn.Cache and run cachetest.Run.