Understand Query Credits and Execution Time
Cost controls help you decide how much data to analyze. Query usage records help you understand the work a run actually consumed.
zinzu records Snowflake credits attributed to a tracked query run and how long its processing took. Execution time is available through the query’s execution statistics. The backend also makes recorded credit details available for reporting.
What Do These Numbers Mean?
| Measure | Plain-language meaning |
|---|---|
| Credits utilized | Snowflake’s reported compute credits attributed to the tracked run, including its related processing statements and query acceleration where reported. |
| Execution duration | Elapsed time for the query’s processing call, recorded in milliseconds. It is not the duration of the business actions found in your data. |
| Warehouse size | The size of the Snowflake compute resource used for the run, when available. |
Example: A query may take 12 seconds to execute and report 0.02 credits, while finding customer sequences spanning 30 minutes. These describe three different things: processing time, attributed compute usage, and time between customer actions.
The example figures are illustrative, not a price estimate.
Why Credits May Not Be Available Immediately
A query can finish before its credit information is available. The cost-reporting process waits until at least eight hours after completion before filling in credits from Snowflake’s attribution history. Scheduled refresh timing and data availability can add further delay.
A missing credit value means the amount has not been reported yet. It does not mean the query used zero credits.
Use This Information
Compare similar runs to understand the effect of changing the selected date range or sampling percentage. Consider both credits and execution time: a faster run does not necessarily use fewer credits.
Important Notes
- Credits are usage units, not a dollar amount. Their monetary value depends on your Snowflake pricing.
- The recorded amount covers attributed query compute and query acceleration where reported. It is not the total bill for storage, idle compute, app hosting, or separate AI requests.
- Execution duration measures processing, not time spent waiting in zinzu’s submission queue.
- Usage is tracked per execution attempt; a retry can have a separate record.