Frontend Realtime Sync

Organization-scoped realtime synchronization pattern that automatically invalidates TanStack Query caches when database changes occur.

Core Concepts

Realtime Table Events System

Organization-scoped notification system that eliminates need for organization_id on junction tables:

• •Backend: Triggers insert events into realtime_table_events table
• •Frontend: Subscribes to realtime_table_events filtered by organization
• •Event payload: { table_name, event_type, organization_id }
• •No hardcoded table list - works for any table with backend trigger

Channel Pattern

• •Separate channel per organization: org:org_123
• •Subscribes to realtime_table_events table with filter: organization_id=eq.{orgId}
• •Automatically resubscribes when user's organization memberships change

Query Invalidation (Granular Record-Level)

• •Backend trigger inserts notification into realtime_table_events with table_name and record_id
• •Frontend receives INSERT event with both table_name and record_id in payload
• •Invalidation logic:
  • •.list() queries: ALWAYS invalidate on any table change
  • •.record(id) queries: ONLY invalidate when record_id matches
• •TanStack Query automatically re-fetches affected queries
• •UI updates in real-time across all users in the organization

Performance benefit: With 500 concurrent users updating different records, only affected queries refetch instead of all table queries.

Architecture

Database Change (Postgres)
    |
    v
Trigger: get_organization_id_for_change()
(Traces change to affected organization + extract record ID)
    |
    v
INSERT into realtime_table_events
(table_name, event_type, organization_id, record_id)
    |
    v
Supabase Realtime Publication
    |
    v
Organization Channel (org:org_123)
    |
    v
useSupabaseRealtimeSync Hook
    |
    v
Extract table_name and record_id from payload
    |
    v
Query Invalidation (granular predicate):
  - .list() queries: Always invalidate
  - .record(id) queries: Only if record_id matches
    |
    v
UI Re-fetch & Update (only affected queries)

Adding New Tables to Realtime

Frontend: No changes needed! Just ensure QueryKeys use .list() for list queries or .record(id) for single-record queries.

Backend only (in migration):

1. •

   Create your table (no organization_id required for junction tables)

   sql

   -- Example: Junction table without organization_id
   CREATE TABLE my_junction_table (
     parent_id uuid REFERENCES parent_table(id),
     child_id uuid REFERENCES child_table(id),
     PRIMARY KEY (parent_id, child_id)
   );
   

2. •

   Update get_organization_id_for_change() function:

   sql

   CREATE OR REPLACE FUNCTION public.get_organization_id_for_change()
   RETURNS TEXT AS $$
   DECLARE
     org_id TEXT;
     record_data JSONB;
   BEGIN
     IF TG_OP = 'DELETE' THEN
       record_data := to_jsonb(OLD);
     ELSE
       record_data := to_jsonb(NEW);
     END IF;
   
     CASE TG_TABLE_NAME
       -- Add your new table:
       WHEN 'my_junction_table' THEN
         SELECT p.organization_id INTO org_id
         FROM parent_table p
         WHERE p.id = (record_data->>'parent_id');
   
       -- ... other tables ...
     END CASE;
   
     RETURN org_id;
   END;
   $$ LANGUAGE plpgsql SECURITY DEFINER SET search_path = public;
   

3. •

   Attach trigger:

   sql

   DROP TRIGGER IF EXISTS trigger_notify_org_change ON my_junction_table;
   CREATE TRIGGER trigger_notify_org_change
     AFTER INSERT OR UPDATE OR DELETE ON my_junction_table
     FOR EACH ROW
     EXECUTE FUNCTION notify_organization_of_table_change();
   

4. •

   Verify QueryKeys exist (usually already present):

   typescript

   // src/utils/query/queryKeys.ts
   export const QueryKeys = {
       my_junction_table: createTableFactory("my_junction_table"),
   };
   

That's it! Frontend automatically invalidates queries when it receives table_name: "my_junction_table" in event payload.

Global Tables (No Realtime)

Tables WITHOUT organization_id should use manual invalidation:

• •organizations - Use manual invalidation in mutation hooks
• •profiles - Use manual invalidation in mutation hooks
• •role_permissions - Use manual invalidation in mutation hooks

Why no realtime for global tables?

• •No way to scope by organization
• •Would broadcast to all users (privacy/performance issue)
• •Manual invalidation provides instant UI feedback for mutations

Provider Setup

// App.tsx
<Provider_Query>
  <Provider_ANTD>
    <Provider_SupabaseRealtimeSync>
      <RouterProvider router={router} />
    </Provider_SupabaseRealtimeSync>
  </Provider_ANTD>
</Provider_Query>

The provider should be:

• •After QueryClientProvider (needs query client context)
• •Inside authentication providers (needs user context)
• •Before router/app components

Realtime Flow for Org-Scoped Tables

1. •User A creates a project in organization org_123
2. •Postgres INSERT trigger fires on projects table
3. •Trigger calls get_organization_id_for_change() → returns org_123
4. •Trigger inserts into realtime_table_events: {organization_id: 'org_123', table_name: 'projects', event_type: 'INSERT'}
5. •Supabase Realtime broadcasts INSERT event to channel org:org_123
6. •All users subscribed to org:org_123 receive notification event
7. •Frontend extracts table_name: 'projects' from payload
8. •handleTableChange invalidates all queries with 'projects' in query keys
9. •TanStack Query re-fetches affected queries
10. •UI updates automatically for all users in the organization

Key Advantage: Works even for junction tables without organization_id column, as backend traces relationship to affected organization.