Caching Apache Superset with Redis
Configuring Superset's data, metadata, filter state and Explore caches on Redis: cache timeouts per chart and dataset, result size limits, Redis Sentinel and what happens when Redis goes down.
Apache Superset performance and scaling
Caching, worker, web server and dashboard design choices that decide how fast Superset feels and how much load it puts on the warehouse.
Speeding up slow Apache Superset dashboards
Finding why a dashboard is slow and fixing it: chart count and tabs, virtualization and deferred data fetch, cache hit rates, warehouse query plans and pre-aggregated datasets.
Sizing Gunicorn workers for Apache Superset
Choosing worker class, worker count, threads and timeouts for Superset's web tier, the BigQuery gevent caveat, and scaling web replicas behind a load balancer.
Scaling Apache Superset Celery workers
Sizing Celery workers for async queries, reports, thumbnails and cache warmup: queues per workload, concurrency, a single beat scheduler, broker choice and autoscaling on Kubernetes.
Warming the Apache Superset cache before users arrive
Pre-computing popular dashboards with Celery beat cache warmup tasks: strategies, the executor user, warming native filter options and the warehouse cost of warming too much.
Dashboard and chart thumbnails in Apache Superset
Enabling thumbnails for dashboard and chart lists: the feature flag, Celery and headless browser requirements, the thumbnail cache, ETags and the load thumbnails add to workers.
Fixing query timeouts in Apache Superset
Diagnosing timeouts at each layer: web server and proxy timeouts, SQL Lab async limits, chart query timeouts, warehouse limits and moving long queries to async execution.
Controlling warehouse query cost from Apache Superset
Keeping Snowflake, BigQuery, Redshift and Athena bills in check when dashboards fan out queries: caching, row limits, auto-refresh limits, deferred loading and pre-aggregation.
Sizing an Apache Superset deployment
Estimating web, worker, cache and metadata database resources from user counts and usage peaks rather than data volume, with load spikes from shared dashboards and room for reports and thumbnails.
