Exam Focus · Part 4
Files wait in stages until COPY or Snowpipe loads them. Each path has its own trigger, compute and defaults.
Know these cold
| Topic | Remember |
|---|---|
| Stages | Internal (@~, @%t, named) take PUT/ |
| Format precedence | COPY > stage > table. None set = CSV. Format options don't merge |
| Load history | Bulk COPY: 64 days, on the table. Snowpipe: 14 days, in the pipe |
| ON_ | Bulk COPY: ABORT_ |
| COPY transforms | Reorder, omit, cast columns. No WHERE, JOIN or GROUP BY |
| Check a load | VALIDATION_: dry run before. VALIDATE: errors after |
| Snowpipe vs Streaming | Snowpipe: files, auto-ingest event or REST, serverless, per GB. Streaming: rows, seconds |
| Sizes | Load files 100–250 MB compressed. VARIANT value up to 128 MB |
| External vs directory table | External: query files in place, read-only, no Time Travel. Directory: file catalog |
| Constraints | Standard tables enforce only NOT NULL and CHECK. PK, UNIQUE, FK don't block loads |
Traps
- Myth Snowflake picks the format from the file extension.Fact It never guesses the type.
- Myth Rerunning COPY loads the same files again.Fact Load metadata skips them. FORCE = TRUE reloads, with duplicates.
- Myth Cloud keys in CREATE STAGE are the recommended setup.Fact Use a storage integration: no secrets in SQL.
Before you turn the page
- For Snowpipe and Streaming, name the trigger, compute and billing.
- Explain why a rerun COPY loads zero rows, and how to force a reload.
- Contrast external and directory tables in one sentence each.

