Skip to content

multi-tenant.md

Latest commit

 

History

History
139 lines (93 loc) · 5.05 KB

multi-tenant.md

File metadata and controls

139 lines (93 loc) · 5.05 KB

Multi-Tenant Apps

Sigra now ships logical multi-tenancy through organizations. The model is row-based: one database, shared tables, explicit organization_id scoping, and a current scope that carries the active organization and membership for the signed-in user.

This guide is about the posture Sigra actually ships. It does not ask you to build PG schema-per-tenant around Sigra itself.

The shipped model: logical multi-tenancy

The default install gives you:

  • organizations
  • organization_memberships
  • organization_invitations
  • active_organization_id on the session row
  • %Scope{active_organization: ..., membership: ...}

The working rule is simple: data owned by an organization gets an organization_id column, and application queries must scope through the active organization before they hit the repo.

Scope queries with for_org/2

Sigra.Organizations.Query.for_org/2 is the primary query helper:

def list_projects(scope) do
  Project
  |> Sigra.Organizations.Query.for_org(scope)
  |> Repo.all()
end

It also accepts a raw organization ID:

def list_projects_for(org_id) do
  Project
  |> Sigra.Organizations.Query.for_org(org_id)
  |> Repo.all()
end

Use it on every org-owned schema. If a schema does not have an organization_id field, for_org/2 raises immediately instead of silently pretending the query is safe.

What "logical multi-tenancy" means in Sigra

Logical multi-tenancy means:

  • one Postgres database
  • one shared schema for Sigra's auth tables
  • org ownership represented by rows and foreign keys
  • per-request access controlled by membership plus scoped queries

That is the default and recommended posture for Sigra-powered SaaS apps. It keeps the generated auth system, audit trail, and sessions in one coherent runtime model.

Why Sigra rejects PG schema-per-tenant for itself

PG schema-per-tenant can be a valid architecture for some products, but Sigra does not treat it as its own multi-tenant primitive.

Reasons:

  • the generated auth surface assumes one shared auth schema
  • per-tenant schema migration orchestration is operationally heavier
  • cross-org membership, invitations, passkeys, and audit workflows are simpler in a shared-schema model
  • a missed search_path or prefix edge can create confusing partial isolation

In other words: Sigra supports logical multi-tenancy directly. If your product later needs PG schema-per-tenant for app-specific data, layer that on deliberately in your own code instead of expecting Sigra to run that model for auth.

Add organization_id to your own schemas

For app-owned tenant data, add an organization_id foreign key and index it:

def change do
  create table(:projects, primary_key: false) do
    add :id, :binary_id, primary_key: true
    add :organization_id, references(:organizations, type: :binary_id), null: false
    add :name, :string, null: false

    timestamps(type: :utc_datetime)
  end

  create index(:projects, [:organization_id])
end

Then keep repo access behind scope-aware functions:

def create_project(scope, attrs) do
  %Project{}
  |> Project.changeset(Map.put(attrs, :organization_id, scope.active_organization.id))
  |> Repo.insert()
end

def list_projects(scope) do
  Project
  |> Sigra.Organizations.Query.for_org(scope)
  |> Repo.all()
end

Membership is authorization, for_org/2 is data isolation

Keep those jobs separate:

  • membership and role checks decide who is allowed to act
  • for_org/2 decides which rows a query can see

Do not rely on controller params, slugs, or UI state alone. The active organization in scope should drive both authorization and query scoping.

Audit and session implications

Sigra's org-aware runtime can carry active_organization_id in the session and attach organization_id to audit metadata. That gives you a consistent tenant story across:

  • browser sessions
  • organization switching
  • invitation acceptance
  • org-scoped audit review

Keep your app-owned audit queries aligned with the same org boundary.

Testing

Add direct regression around the scoping helper:

test "for_org/2 only returns rows for the active organization" do
  org_a = organization_fixture()
  org_b = organization_fixture()
  scope = scope_fixture(active_organization: org_a)

  project_fixture(organization_id: org_a.id, name: "visible")
  project_fixture(organization_id: org_b.id, name: "hidden")

  projects =
    Project
    |> Sigra.Organizations.Query.for_org(scope)
    |> Repo.all()

  assert Enum.map(projects, & &1.name) == ["visible"]
end

Also keep route-level tests for membership requirements. Query helpers make data isolation shorter to express; they do not replace full request coverage.

Related

  • Getting Started — the default organizations and passkeys walkthrough.
  • Passkeys — passkey-primary login and recovery posture inside an org-aware app.
  • Testing Auth Flows — fixtures and helpers for auth-heavy integration tests.