Scaling GraphQL Subscriptions for Real-Time Dashboards

In this talk, I'm presenting how GraphQL Subscriptions helped to build the real-time dashboards and how we scaled the implementation of GraphQL Subscriptions

Transcript

1. GRAPHQL SUBSCRIPTIONS AT SCALE FOR REAL-TIME DASHBOARD
2. VIMALRAJ SELVAM email2vimalraj
3. Source: https://www.kisspng.com/png-social-media-command-center-computer-icons-command-2009405/preview.html
4. AGENDA ▸ Real time dashboard? ▸ Grafana and its pull-based model ▸ Need for push-based model ▸ GraphQL Subscriptions at rescue ▸ Highly Scalable GraphQL Subscription Architecture
5. REAL TIME DASHBOARD Source: https://play.grafana.org/d/000000012/grafana-play-home
6. DEMO: PULL BASED MODEL
7. PULL BASED MODEL (CONTD…) Query every 5s ▸ 1 Dashboard = 10 Panels ▸ Refresh rate = 5 seconds ▸ 100 users x 10 Panels = 1000 req. per 5s ▸ 1 request = 10 KB => 1000 requests = 10 MB WHAT WE NEED THEN? Push Data to Client => Web Sockets
8. GRAPHQL SUBSCRIPTIONS ▸ Server sends data to subscribed clients when some event occurs ▸ Over the Websockets DEMO
9. HIGHLY SCALABLE GRAPHQL SUBSCRIPTIONS ARCHITECTURE LOAD BALANCER GraphQL Subscriptions Server Query Every 5s Websocket ▸ 1 Dashboard = 10 Panels ▸ 3 servers = 3 req. / 5s ▸ 100 users x 10 Panels = 3 req. / 5s ▸ 1 request = 10 KB => 3 requests = 30 KB Calls to Data sources are redundant
10. HIGHLY SCALABLE GRAPHQL SUBSCRIPTIONS ARCHITECTURE LOAD BALANCER GraphQL Subscriptions Server Query Every 5s Websocket Worker PUBLISH SUBSCRIBE ▸ 1 Dashboard = 10 Panels ▸ 3 servers = 1 req. / 5s ▸ 100 users x 10 Panels = 1 req. / 5s ▸ 1 request = 10 KB
11. HIGHLY SCALABLE GRAPHQL SUBSCRIPTIONS ARCHITECTURE Worker Master Worker Worker Slave Slave Register as a Master with TTL Compete for Master
12. THANK YOU email2vimalraj