Performance and Cost
Understand analysis scope, sampling, credits and execution time.
37. How can I explore a large dataset without analyzing every owner?
Choose a relevant date range and a lower sampling percentage before submitting. Inspect the patterns, refine the query, and increase coverage when needed.
If data skew protection is configured, it can also exclude exceptionally large owner histories. These reduce processing work but do not guarantee a fixed cost reduction.
38. Does sampling select owners or individual records?
It selects owners consistently by their identifier. Eligible records for selected owners remain available across the chosen sources and query range.
10% sampling does not keep only one in ten actions from each owner, which would break sequences. The number of selected owners is approximate.
39. Can sampling miss rare behavior?
Yes. A rare pattern may belong only to owners outside the sample. No match in a small sample does not prove the behavior is absent.
Use 100% for all eligible-owner coverage when validating rare or important behavior. Counts are based on the analyzed population, not automatically scaled to full-population estimates.
40. What is data skew, and why might an owner be excluded?
Data skew means a few owners have far larger histories than others and can dominate processing.
When enabled, protection counts an owner’s source records across selected sources within the query range. Owners exceeding the configured threshold are excluded entirely for that run; exactly the threshold is allowed.
Protection applies only when enabled for the run. Review exclusions carefully if unusually heavy activity is relevant to your analysis.
41. How is the sequence record limit different from the data skew threshold?
The sequence limit stops one matching attempt after a configured number of records. An owner can still have other sequences.
The data skew threshold checks total owner history in the query range and can exclude the entire owner before sequence matching.
Example: sequence limit 50 controls each attempt; skew threshold 50,000 excludes owners above 50,000 source records. The latter is an example configuration, not a universal default.
42. What do execution time and query credits tell me?
My Queries provides execution statistics, including execution time and warehouse size when available.
Tracked query-credit information reports attributed compute usage, including query acceleration where reported. Credits are usage units, not dollars or the full Snowflake bill. Separate AI requests and other account costs are not included in that query-compute total.
43. Why are credits unavailable immediately after a query finishes?
Credit reporting depends on Snowflake attribution history. Credit information is collected no earlier than eight hours after completion, and reporting schedules or data availability can add delay.
Missing credits mean “not yet reported,” not zero consumption. Duration information may be available sooner.
44. Can I cancel a running query?
Yes. My Queries includes cancellation for processing queries and submits a cancellation request.
Cancellation does not erase compute already consumed or guarantee every operation stops instantly. Check the resulting status before treating cancellation as complete.