It’s an impressive job. Recently, I’ve been trying to work with time series graphs in an environment with limited capabilities (a Dash server running on a Raspberry Pi 3). I have the data stored in a database (SQLite), and I think the main bottleneck occurs when loading the data for visualization. Perhaps a similar implementation could be done, but one that avoids loading all the data into memory?
They don’t have the budget for a Raspberry pi 5?
@ulibussi what you are describing is exactly what we do in flexviz: GitHub - flex-analytics/flexviz: Interactive cross-filter dashboards on 100M+ rows, in pure Python. Polars-native · GitHub
In flexviz we keep the data lazy (i.e., not in memory) and stream the data from disk to ram → allows the visualize larger-than-ram dataset. Basically, this is the successor of the plotly-resampler library; in addition to line charts, it supports 9 other trace types (through highly optimized data aggregation), the server is now stateless, and it comes with native cross-filtering.