Panel: The Dashboard That Grew - A Scaling Saga

on YouTube ↗30m

" summary

Claims a migration from Streamlit to Panel to handle scale and shows the concrete numbers: 100 to 2000+ users in less than 2 months. Describes running on Kubernetes, hits performance pain: cascading filter logic, data growth forcing caching, tabs causing full-graph renders, and a critical bug where state is shared across sessions because of how apps are served via Bokeh/websockets. Explains fixes tried, tradeoffs between Python-first UI vs separating an API/backend, and which choices actually worked.

AI-written summary · about

" chapters (12)

1. 0:00Introduction
2. 1:08What is the dashboard and why it grew too much
3. 4:41Streamlit’s strengths and limits
4. 7:52Panel introduction
5. 10:21Panel features deep dive
6. 13:56Filter cascading issues
7. 16:43Data loop trade-offs
8. 20:37Shared state incident
9. 22:01Tabs performance pitfalls
10. 24:51Testing scaling bottlenecks
11. 26:14Key lessons and future
12. 27:39Q&A

" related talks

• 50mPanel: Large Scale Streaming in ProductionMonal Daxini, Maxim Fateev, Ted Dunning, Xiaowei Jiang, Jamie Grier · Flink Forward Berlin 2016
• 48mPanelTaylor Barnett, Katy Farmer, Kat Cosgrove, Arathi Ramani · DevOpsDays Seattle 2021
• 40mHacking the Pachyderm: Scaling Servers and PeopleHazel Weakly, Preston Doster · SREcon23 Americas
• 46mPanel Discussion: The number of orcheBryan Hunt, Verocina Lopez, David Schainker, Thomas Depierre, Jani Leppanen, Jake Morrison · Code Mesh LDN 2020
• 42mThe Challenges Managing a Kubernetes-Based Machine Learning InfrastructureEd Shee, Yuzhui Liu, Keith Laban, Keshi Dai · KubeCon + CloudNativeCon North America 2022
• 47mScaling: from 0 to 25 million usersJosip Stuhli · Open Source Monitoring Conference 2024
• 31mKeynote Panel: Node.js at Massive ScaleY. Xiao, J. Chan, M. Hiskes, A. Slaton, K. Borchers · Node+JS Interactive 2018
• 33mA Monolith, a Microservices, and a Function Walk into a BarIan Crosby · DevOpsDays Boston 2019