Study SnowPro DEA-C02 Streams and CDC: key concepts, common traps, and exam decision cues.
When Snowflake asks about near real-time CDC, the first job is to separate the object that captures change from the object that runs the work.
| Requirement | Better first instinct |
|---|---|
| track row-level changes from an upstream table or view | stream |
| run logic on a schedule or trigger boundary | task |
| build explicit CDC pipeline behavior | stream plus task pattern |
| Concern | Stronger first answer |
|---|---|
| expose the changes | stream |
| run downstream work | task |
| build the full explicit CDC path | stream + task |
This is the core distinction in the lesson, and Snowflake often tests it directly.
| If the stem says… | Strong reading |
|---|---|
| “capture changes” | stream boundary |
| “execute downstream transformation” | task or orchestration boundary |
| “near real-time pipeline” | decide whether explicit CDC plus tasking is needed or a managed path fits better |
Candidates often want one object to own everything. Snowflake does not work that way here:
The strongest answer usually preserves those responsibilities instead of collapsing them into one fuzzy pipeline noun.
| Trap | Better rule |
|---|---|
| using tasks to solve change capture directly | tasks run work; they do not capture change by themselves |
| using streams as if they schedule execution | streams expose changes; they do not own orchestration |
| collapsing the full pipeline into one object | near real-time systems still have layered responsibilities |
| Scenario clue | Stronger answer shape |
|---|---|
| “need row-level changes from source” | stream |
| “need downstream logic to run on a schedule or trigger” | task |
| “need explicit CDC pipeline behavior” | stream + task |
| “question blurs capture and execution” | separate them before choosing |
Near-real-time CDC questions usually hinge on separating capture from execution. Streams expose row-level changes. Tasks run downstream work on a schedule or trigger boundary. The weak answer usually assumes a task can detect changes by itself when the actual requirement is first to surface change data.