Lesson 17 · Senior CDC practice

Reviewing CDC and streaming-ETL designs like a senior engineer

How to judge whether a CDC pipeline, connector fleet, or streaming-ETL design is truthful about delivery, backfills, schema safety, and operational trade-offs.

Your win: review CDC and streaming-ETL designs for hidden trade-offs, weak assumptions, and overclaimed guarantees instead of stopping at “we use Debezium and Connect.”

Why review is its own skill

By the time you reach this lesson, you already know the moving parts. You know what Debezium does, what the connectors do, what hephaestus does, and how the warehouse path is shaped. The new step is different: can you tell whether a proposal using all those words is actually good?

That is why review is its own skill. A design can name all the right components and still be weak, incomplete, or misleading about what the system truly guarantees.

In plain English Plain English: senior review asks what the pipeline really guarantees, what it does not guarantee, and where it is quietly fragile.

The questions strong reviews keep returning to

The strongest earlier lessons keep circling back to consequence. This review page should do the same. If someone says “CDC,” what happens under replay? If they say “backfill,” what happens to slot continuity? If they say “the generator handles it,” what happens to meaning drift? If they say “auto-heal,” what happens to poison-record evidence?

Those are not nitpicks. They are the real review questions. They turn platform vocabulary into engineering judgment.

The review question What exact guarantees does this pipeline provide about capture continuity, replay safety, schema compatibility, and failure handling — and what important trade-offs is it quietly hiding?

Where strong review comments usually land

claims that sound like exactly-once when the design is really at-least-once plus idempotent upsert
backfill plans that talk about snapshots but ignore slot continuity and source recreation risk
schema-evolution stories that stop at “the generator handles it” without checking semantic drift
failure-handling flows that mention restartability but never explain poison-record evidence or no-DLQ trade-offs
warehouse or transform paths that quietly change the meaning of the data without being reviewed as contract changes

Backend use case This lesson helps you review repo CDC changes with sharper questions about IncrementalSnapshot, hephaestus reconcile behavior, schema intersections, ksqlDB-owned connectors, no-DLQ sinks, and chained CDC into lmsdwh.

Common mistake Letting a CDC design sound stronger than it is because it uses the right platform vocabulary.

Use the repo map as the review anchor

The strongest CDC reviews in this repo start by checking claims against the actual source, sink, and control-plane behavior captured in the map.

→ Repo CDC map
→ docs/cdc/lessons/0012-operating-the-pipeline.html

Check yourself (from memory)

Q1. What is the most senior review question for a CDC design?

Senior review is about guarantee truthfulness, not buzzword coverage.

What should a senior engineer ask when reviewing a CDC pipeline design?

recall, then click to reveal

What does this pipeline really guarantee about continuity, replay, schema safety, and failure handling, and where are the important trade-offs or fragilities still unresolved?

Want a compact PR-review checklist for CDC-backed designs in this repo? Ask me.

Sources. Repo CDC map; existing CDC lessons; official docs.