Orchestrating Data Pipelines at Scale with Temporal
summary
Fivetran runs over 5 million daily syncs on Temporal workflows, migrates from regionally fragmented schedulers to a worker-pool model, and cuts job startup latency from 25s to 5s. The migration uses Temporal Cloud, a custom operator for long‑running jobs and bin‑packing, and a custom gRPC gateway for hybrid deployments that authenticates agents and routes requests by namespace. Metrics tracked include schedule‑start latency, task depth, worker utilization and cost dashboards; gradual rollout and namespace splitting resolve non‑determinism and scaling issues.
AI-written summary · about
chapters (12)
- 0:00Introduction to talk
- 1:06Fiverr’s data platform overview
- 7:15Why Fiverr adopted Temporal
- 9:48Moving from legacy schedulers
- 17:12Implementation phases of Temporal
- 21:01Pipeline workflow design
- 27:21Handling complex pipelines
- 30:53Metrics and monitoring during migration
- 32:19Challenges and lessons learned
- 35:10Hybrid deployment with Temporal
- 38:40Future plans and next steps
- 40:08Q&A session
related talks
30mOrchestrate Your Business Workflows with Temporal
26mTemporal: A Better Way to Build Modern Applications
42mUsing Temporal Workflows for Seamless Dev + Research Collaboration & Continuous Algo D
23mDesigning a Workflow Engine From First Principles
26mOperating Cassandra in Kubernetes with Temporal
19mGeorge Fraser, Fivetran | Databricks Data + AI Summit 2024
26mMicroservices Madness: Application And Infrastructure Automation Using Temporal
16mAnjan Kundavaram, Fivetran | Databricks Data + AI Summit 2024
