Sample questions
Streaming FundamentalsDifficulty 1
What best describes batch processing in data systems?
- aData is collected over a period of time and processed together as a group at scheduled intervals✓
- bData is processed continuously as each individual record arrives, with results available within milliseconds of each event
- cData is processed only when a downstream consumer explicitly requests it, one record at a time
- dData is discarded immediately after processing so no history is retained
Explanation:Batch processing accumulates data over a window and processes it together at scheduled intervals, trading immediacy for higher throughput and simpler handling. Option b actually describes stream processing. Option c describes an on-demand pull pattern, not batch. Option d confuses the processing model with a data retention policy.
Streaming FundamentalsDifficulty 1
What best describes stream processing in data systems?
- aData is grouped into large historical batches and processed once per day regardless of when it was generated
- bData is processed continuously, record by record, as it arrives, with low latency✓
- cData can only be processed after being fully loaded into a single data warehouse table
- dData is processed exactly once per week on a fixed schedule
Explanation:Stream processing handles data continuously as it arrives, in contrast to batch's fixed-schedule grouping. Options b and d describe batch-like fixed schedules, not streaming. Option c describes a load-then-query pattern rather than continuous processing.
Streaming FundamentalsDifficulty 2
A retail company needs to generate a report of yesterday's total sales by region, delivered each morning to the finance team. Which processing approach fits this need best?
- aNeither approach applies, since financial reports must be generated manually
- bStream processing, since finance always requires millisecond-level freshness even for daily reports
- cBatch processing, since the report reflects a finalized day and can tolerate hours of delay✓
- dStream processing, since batch systems cannot aggregate data by region
Explanation:Since the report only needs to reflect a finalized day and tolerates hours of delay until the next morning, batch processing is the simpler, sufficient choice. Option b demands unnecessary freshness the use case doesn't need. Option d is factually wrong — batch systems routinely aggregate by any dimension, including region.
Streaming FundamentalsDifficulty 2
An e-commerce platform wants to flag a payment as potentially fraudulent within a couple of seconds of the transaction happening, before the payment is approved. Which processing approach fits this need best?
- aEither approach is equally suitable, since latency does not matter for fraud detection
- bBatch processing, since a nightly job can catch fraud early enough for most cases
- cBatch processing, since fraud patterns only emerge after collecting a full month of data
- dStream processing, since decisions must be made instantly as transactions arrive✓
Explanation:Fraud detection that must act within seconds, before approval, needs continuous, low-latency processing as events happen — that is stream processing. Options b and c both wrongly assume a delay of hours or a month is acceptable, which defeats the purpose of catching fraud before the payment is approved.
Streaming FundamentalsDifficulty 1
In stream processing, what does 'at-least-once' delivery semantics guarantee?
- aEvery record will be processed one or more times, but no record will be silently dropped✓
- bEvery record is guaranteed to be processed exactly one time, with no duplicates and no omissions
- cSome records may be silently dropped, but no record will ever be processed more than once
- dRecords are only processed if the consumer explicitly polls for them within a fixed time limit
Explanation:At-least-once guarantees no silent loss but allows duplicates, since a record may be redelivered until it is acknowledged as processed. Option b describes exactly-once semantics. Option c describes at-most-once (loss allowed, duplicates not).
Streaming FundamentalsDifficulty 2
In stream processing, what does 'at-most-once' delivery semantics guarantee?
- aEach record is delivered one or more times, and duplicates are always eliminated automatically by the system
- bEach record is delivered zero or one times; it may be lost, but it will never be processed more than once✓
- cEach record is guaranteed to be delivered and processed precisely one time, no exceptions
- dRecords are held until every downstream consumer confirms processing before being removed
Explanation:At-most-once prioritizes never duplicating over never losing data — a record might simply not arrive at all. Option b describes at-least-once semantics. Option c describes exactly-once semantics.