English for Snowflake Developers

Snowflake’s separation of storage and compute means its English vocabulary is as much about cost and concurrency as it is about SQL. A team discussing “the warehouse” might mean the compute cluster, not the data — and getting that distinction wrong in a conversation about cost can send an entire investigation in the wrong direction. This guide covers the English used when discussing Snowflake work with a team.

Key Vocabulary

Virtual warehouse — a cluster of compute resources you spin up to run queries, billed by the second while running, independent of where the data is stored. “Let’s scale the ETL warehouse up to a Large for this load, and scale it back down afterward so we’re not paying for idle compute.”

Micro-partition — the small, immutable, automatically-managed chunks Snowflake stores table data in, each with metadata used for pruning during query planning. “The query is scanning far more micro-partitions than expected — that usually means the clustering key isn’t aligned with our filter column.”

Clustering key — a column or expression used to co-locate related rows within micro-partitions, improving pruning for queries that filter on it. “We should add a clustering key on event_date — right now every query filtering by date has to scan partitions across the whole table.”

Time travel — a feature letting you query, clone, or restore a table as it existed at an earlier point in time, within a configured retention window. “Before we run the backfill, let’s confirm time travel is enabled with enough retention that we can roll back if the numbers look wrong.”

Zero-copy clone — creating a full logical copy of a table or database that shares the underlying storage until either copy is modified, making it nearly instant and free at creation time. “Instead of exporting and reloading a full copy for testing, just zero-copy clone the production schema into a dev database.”

Warehouse auto-suspend — a setting that automatically pauses a virtual warehouse after a period of inactivity, stopping billing until the next query resumes it. “Set auto-suspend to sixty seconds on that ad hoc warehouse — it’s been left running idle overnight and racking up cost.”

Common Phrases

• “Which warehouse is this query running on, and is it sized appropriately for the workload?”
• “Are we scanning a disproportionate number of micro-partitions here — should we revisit the clustering key?”
• “Can we zero-copy clone this schema for the test environment instead of re-ingesting the data?”
• “Is time travel retention long enough on this table to recover from a bad load?”
• “Is auto-suspend configured on this warehouse, or is it running idle between jobs?”

Example Sentences

Reviewing a cost anomaly: “The spike traces back to a warehouse someone spun up as X-Large for a one-off backfill and never scaled back down — let’s add an auto-suspend policy so this doesn’t recur.”

Explaining a design decision: “We chose a clustering key on the tenant ID because nearly every downstream query filters by tenant, so pruning on that column gives us the biggest scan reduction.”

Describing an incident: “We used time travel to restore the table to its state before the faulty merge statement ran, which let us recover without needing last night’s backup.”

Professional Tips

• Say “warehouse” to mean compute, not data — in Snowflake conversations this is a common source of confusion for newcomers from traditional data warehouse backgrounds.
• When discussing slow queries, ask “how many partitions did this scan versus how many did it need to?” — this is the standard framing for a pruning problem in English-speaking data teams.
• Use “zero-copy clone” precisely — it signals you understand the storage-sharing behavior, not just “we made a copy.”
• Distinguish “auto-suspend” (pausing compute) from “auto-resume” (restarting it on the next query) when explaining warehouse cost behavior.

Practice Exercise

1. Explain in two sentences why a poorly chosen clustering key increases query cost.
2. Write a one-sentence recommendation for reducing cost on a warehouse left running idle.
3. Describe, in your own words, when you would use a zero-copy clone instead of a full data export.