Stream Processing: Windows, Watermarks and Late Events · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

Cut the endless stream into windows

Computing · Data & Databases · ages 23-24
Name ______________________   Date ____________

  1. What is event time?

    • When the system happened to see the record
    • When the record happened in the world
    • When the window finished closing

  2. What is a window over a stream?

    • A fixed slice of time collecting its own result
    • The final end of the stream
    • A faster network cable

  3. What does a watermark claim?

    • That event time has passed a point
    • That no record will ever arrive late
    • That windows are no longer needed

  4. What does holding a watermark back buy, and what does it cost?

    • Faster answers with no downside
    • More complete windows at the price of slower answers
    • Freedom from windows entirely

  5. A phone's taps arrive ten minutes late. Which window gets them under event time?

    • The window when they arrived
    • A brand-new window each time
    • The window when they happened

  6. A public leaderboard must never revise past numbers. Which late policy fits?

    • Fold late records in with corrections
    • Wait forever for every record
    • Drop late records and say so

  7. Folding late records in keeps every past answer final.

    Circle one:   True   False

  8. A team holds its watermark far back yet promises instant and complete answers. What is wrong?

    • Late records vanish once a watermark exists
    • Holding back only raises cost, never delay
    • Waiting buys completeness with delay, so instant plus complete is impossible
LightMySky · lightmysky.comW1-mt_CfzXL_cL0D-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

Cut the endless stream into windows W1-mt_CfzXL_cL0D-s1

  1. When the record happened in the world · Event time follows the world, not the system's clock.
  2. A fixed slice of time collecting its own result · Each slice gathers its own answer because no end ever arrives.
  3. That event time has passed a point · It marks progress so closed windows can report.
  4. More complete windows at the price of slower answers · Waiting catches late records but delays every answer.
  5. The window when they happened · Event time places records where they belong in the world.
  6. Drop late records and say so · Dropping keeps published numbers final at the cost of misses.
  7. False · Corrections revise the past, so finality is lost.
  8. Waiting buys completeness with delay, so instant plus complete is impossible · Every wait for stragglers is time users feel.
Worksheet · LightMySky