Use stable identifiers, idempotent writes and durable progress, and reconcile uncertain outcomes before repeating side effects.
Illustrative scenario
A target write times out after it may have committed.
Explain your reasoning
Apply the answer to this scenario: What evidence distinguishes failure from unknown completion? State your assumptions, the expected outcome and how you would check it before changing a real system.
Common follow-up
What evidence distinguishes failure from unknown completion?
Capture ordered source changes with initial-snapshot coordination, durable offsets and explicit insert, update and delete semantics.
Illustrative scenario
A customer table feeds an analytical target.
Explain your reasoning
Apply the answer to this scenario: How are gaps between snapshot and streaming avoided? State your assumptions, the expected outcome and how you would check it before changing a real system.
Common follow-up
How are gaps between snapshot and streaming avoided?
It preserves history by closing a previous version and creating a new version with effective boundaries.
Illustrative scenario
A customer moves between regions.
Explain your reasoning
Apply the answer to this scenario: How do you prevent overlapping active periods? State your assumptions, the expected outcome and how you would check it before changing a real system.
Define allowed lateness and correction rules using event time and business requirements.
Illustrative scenario
Yesterday's sale arrives after today's dashboard refresh.
Explain your reasoning
Apply the answer to this scenario: Which outputs need correction? State your assumptions, the expected outcome and how you would check it before changing a real system.
Inspect key distributions and task metrics to find disproportionate work before changing partitions.
Illustrative scenario
One customer key dominates a join.
Explain your reasoning
Apply the answer to this scenario: Would salting preserve the required semantics? State your assumptions, the expected outcome and how you would check it before changing a real system.
Replicating a sufficiently small input can avoid shuffling the large side, but consumes executor memory.
Illustrative scenario
A small lookup joins a large event dataset.
Explain your reasoning
Apply the answer to this scenario: How do you prove the lookup fits safely? State your assumptions, the expected outcome and how you would check it before changing a real system.
More than one source row can match a target row for an update, depending on statement and engine semantics.
Illustrative scenario
A batch includes two versions of an order.
Explain your reasoning
Apply the answer to this scenario: What deterministic rule selects the winner? State your assumptions, the expected outcome and how you would check it before changing a real system.
Restore compatible state and offsets, verify retained source data and reconcile sink effects before resuming.
Illustrative scenario
A streaming job restarts after a failure.
Explain your reasoning
Apply the answer to this scenario: What happens if source retention no longer covers the offset? State your assumptions, the expected outcome and how you would check it before changing a real system.
Common follow-up
What happens if source retention no longer covers the offset?
Classify changes, version contracts and test downstream compatibility rather than accepting every new field blindly.
Illustrative scenario
A producer changes the meaning of a status field.
Explain your reasoning
Apply the answer to this scenario: Can a syntactically compatible change still break consumers? State your assumptions, the expected outcome and how you would check it before changing a real system.
Common follow-up
Can a syntactically compatible change still break consumers?
Read task logs and dependency state, classify transient versus deterministic failure, and verify rerun safety.
Illustrative scenario
A task fails after uploading a file.
Explain your reasoning
Apply the answer to this scenario: Will clearing it produce a second file? State your assumptions, the expected outcome and how you would check it before changing a real system.
Fix the historical range and code version, isolate resources and reconcile outputs without disrupting current loads.
Illustrative scenario
A corrected transformation must rebuild three months.
Explain your reasoning
Apply the answer to this scenario: How are overlaps with live ingestion prevented? State your assumptions, the expected outcome and how you would check it before changing a real system.
Exactly-once is a scoped system guarantee; effectively-once often uses deduplication and idempotent effects across retries.
Illustrative scenario
A processor and external API have different transaction boundaries.
Explain your reasoning
Apply the answer to this scenario: Where does the claimed guarantee end? State your assumptions, the expected outcome and how you would check it before changing a real system.
Isolate deterministic bad input with enough safe context, alert on impact and provide a repair path.
Illustrative scenario
One malformed event repeatedly crashes a consumer.
Explain your reasoning
Apply the answer to this scenario: How do you avoid silently losing business data? State your assumptions, the expected outcome and how you would check it before changing a real system.
Measure resource usage per useful output, then address scanning, shuffles, idle resources and unnecessary recomputation.
Illustrative scenario
A daily job rereads its full history.
Explain your reasoning
Apply the answer to this scenario: What correctness risks come with incremental processing? State your assumptions, the expected outcome and how you would check it before changing a real system.
Common follow-up
What correctness risks come with incremental processing?
Document field semantics and compatibility rules, test producers and consumers and provide migration windows.
Illustrative scenario
A producer adds a new enum value.
Explain your reasoning
Apply the answer to this scenario: Can an old consumer accept it safely? State your assumptions, the expected outcome and how you would check it before changing a real system.
Track the age of the latest relevant business event or completed partition, not only task success.
Illustrative scenario
A job succeeds repeatedly while reading an empty source.
Explain your reasoning
Apply the answer to this scenario: Which freshness SLO reflects user needs? State your assumptions, the expected outcome and how you would check it before changing a real system.
Choose and document physical removal, tombstones or history retention according to the use case and applicable requirements.
Illustrative scenario
A source customer record is deleted.
Explain your reasoning
Apply the answer to this scenario: Which derived datasets must also change? State your assumptions, the expected outcome and how you would check it before changing a real system.
Combine small deterministic fixtures, contract checks, transformation tests and controlled end-to-end reconciliation.
Illustrative scenario
A currency conversion rule changes.
Explain your reasoning
Apply the answer to this scenario: What cases cover rounding and missing rates? State your assumptions, the expected outcome and how you would check it before changing a real system.
Use a managed secret store and scoped identities, and avoid secrets in source, logs and examples.
Illustrative scenario
A scheduled job needs database access.
Explain your reasoning
Apply the answer to this scenario: How will rotation affect running jobs? State your assumptions, the expected outcome and how you would check it before changing a real system.
Connect run identifiers, data-quality outcomes, latency and resource metrics so a business symptom can be traced.
Illustrative scenario
A dashboard is stale while tasks appear green.
Explain your reasoning
Apply the answer to this scenario: Which signal identifies the first broken boundary? State your assumptions, the expected outcome and how you would check it before changing a real system.
Common follow-up
Which signal identifies the first broken boundary?