Optimizing Resources

• Fix an expensive query with layout, not more slots
• Partition a BigQuery table on one date, timestamp, or integer-range column
• Cluster on up to four high-cardinality filter columns to prune blocks
• Use a materialized view, not a logical view, to cut bytes on repeated aggregations
• BigQuery auto-discounts long-term storage after 90 days unmodified
• Pick a Cloud Storage class by access frequency, not by data age alone
• Cold storage classes charge for the full minimum duration on early deletion
• Automate storage tiering with lifecycle rules or Autoclass
• On-demand for unpredictable load, Editions slots for steady analytics
• Commit to capacity only for steady, predictable usage
• Size the reservation baseline to the floor, autoscale for the peak
• Default to per-job ephemeral Dataproc clusters
• Don't lift-and-shift an on-prem Hadoop cluster's persistence model
• Run fault-tolerant batch on Spot VMs for up to ~91% off
• Let Dataflow autoscale workers instead of fixing a worker count
• Autoscale only Dataproc secondary workers, and let graceful decommission outlast the longest job
• Turn on Enhanced Flexibility Mode so preemptible secondary workers can't lose shuffle data
• Lower the worker utilization hint to make Dataflow scale up faster on spikes
• Streaming Engine lets a streaming job scale down to one worker
• Change a running streaming job's worker range with an in-flight update, no restart
• Dataflow Prime Vertical Autoscaling fixes OOM workers without code changes
• Raise numberOfWorkerHarnessThreads for hot keys once workers are maxed out
• On a lifecycle-rule tie, Delete beats SetStorageClass and the coldest class wins

Unlock with Premium — includes all practice exams and the complete study guide.