Exam Bundle

State store size stabilizes as sufficiently old event-time state expires would demonstrate the separate action Set a suitable event-time watermark for the use case, which addresses “A stateful aggregation grows without bound due to no late-event cutoff.” It does not verify the requirement “A stream must deliver at-least-once microbatches to an external system safely” following Use `foreachBatch` with idempotent writes keyed by microbatch ID or business key.

This signal belongs to the adjacent case “A data engineer changes stateful grouping keys and the existing checkpoint is incompatible,” whose remedy is Plan a new checkpoint and controlled state reinitialization or migration. The observed result for the present requirement is Retry of one microbatch does not duplicate the externally committed records.

That observation is a valid indicator for “Two independent streaming queries were configured to share the same checkpoint path” when using Give each logical streaming query its own checkpoint location. The scenario instead tests “A stream must deliver at-least-once microbatches to an external system safely”; the relevant result is Retry of one microbatch does not duplicate the externally committed records.

This action solves a different issue: “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance.” Its normal signal is The batch result has correct grouping without requiring a streaming checkpoint. Here the requested behavior is “Late events within a two-hour tolerance must update event-time window aggregates,” so that signal is not sufficient.

The alternative Use a supported stream-stream join with event-time conditions and watermarks has a distinct purpose—A stateful stream joins two event streams on matching keys and bounded time difference. It would not establish the required outcome (Late events within a two-hour tolerance must update event-time window aggregates); the necessary verification concerns An event arriving within the watermark horizon updates the appropriate open window.

The alternative Inspect streaming progress and state operator late-event metrics alongside watermark settings has a distinct purpose—Operators need to diagnose why late events were discarded from a windowed query. It would not establish the required outcome (Late events within a two-hour tolerance must update event-time window aggregates); the necessary verification concerns An event arriving within the watermark horizon updates the appropriate open window.

The batch result has correct grouping without requiring a streaming checkpoint would demonstrate the separate action Use ordinary batch aggregation instead of relying on streaming watermark state pruning, which addresses “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance.” It does not verify the requirement “Late events within a two-hour tolerance must update event-time window aggregates” following Define a two-hour event-time watermark before stateful window aggregation.

Expired state stops growing with ancient event timestamps would demonstrate the separate action Apply a suitable watermark and check the state cleanup behavior, which addresses “Events beyond the retention watermark should not cause unlimited state accumulation.” It does not verify the requirement “Late events within a two-hour tolerance must update event-time window aggregates” following Define a two-hour event-time watermark before stateful window aggregation.

The observed drop metrics correlate with event timestamps older than the watermark would demonstrate the separate action Inspect streaming progress and state operator late-event metrics alongside watermark settings, which addresses “Operators need to diagnose why late events were discarded from a windowed query.” It does not verify the requirement “Late events within a two-hour tolerance must update event-time window aggregates” following Define a two-hour event-time watermark before stateful window aggregation.

Use a supported stream-stream join with event-time conditions and watermarks belongs to the separate case “A stateful stream joins two event streams on matching keys and bounded time difference,” where Late-data behavior and join state size match the bounded join window would be expected. That differs from the condition in this stem: “Events beyond the retention watermark should not cause unlimited state accumulation.”

This action solves a different issue: “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance.” Its normal signal is The batch result has correct grouping without requiring a streaming checkpoint. Here the requested behavior is “Events beyond the retention watermark should not cause unlimited state accumulation,” so that signal is not sufficient.

This action solves a different issue: “Operators need to diagnose why late events were discarded from a windowed query.” Its normal signal is The observed drop metrics correlate with event timestamps older than the watermark. Here the requested behavior is “Events beyond the retention watermark should not cause unlimited state accumulation,” so that signal is not sufficient.

This signal belongs to the adjacent case “A stateful stream joins two event streams on matching keys and bounded time difference,” whose remedy is Use a supported stream-stream join with event-time conditions and watermarks. The observed result for the present requirement is Expired state stops growing with ancient event timestamps.

This signal belongs to the adjacent case “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance,” whose remedy is Use ordinary batch aggregation instead of relying on streaming watermark state pruning. The observed result for the present requirement is Expired state stops growing with ancient event timestamps.

An event arriving within the watermark horizon updates the appropriate open window pertains to a different control (Define a two-hour event-time watermark before stateful window aggregation) and a different need (Late events within a two-hour tolerance must update event-time window aggregates). It cannot establish that Apply a suitable watermark and check the state cleanup behavior met the stated goal.

Use a supported stream-stream join with event-time conditions and watermarks is appropriate when “A stateful stream joins two event streams on matching keys and bounded time difference”; it would be verified by Late-data behavior and join state size match the bounded join window. It does not directly satisfy “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance,” whose relevant acceptance evidence is The batch result has correct grouping without requiring a streaming checkpoint.

Inspect streaming progress and state operator late-event metrics alongside watermark settings is appropriate when “Operators need to diagnose why late events were discarded from a windowed query”; it would be verified by The observed drop metrics correlate with event timestamps older than the watermark. It does not directly satisfy “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance,” whose relevant acceptance evidence is The batch result has correct grouping without requiring a streaming checkpoint.

Define a two-hour event-time watermark before stateful window aggregation belongs to the separate case “Late events within a two-hour tolerance must update event-time window aggregates,” where An event arriving within the watermark horizon updates the appropriate open window would be expected. That differs from the condition in this stem: “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance.”

Late-data behavior and join state size match the bounded join window pertains to a different control (Use a supported stream-stream join with event-time conditions and watermarks) and a different need (A stateful stream joins two event streams on matching keys and bounded time difference). It cannot establish that Use ordinary batch aggregation instead of relying on streaming watermark state pruning met the stated goal.

That observation is a valid indicator for “Late events within a two-hour tolerance must update event-time window aggregates” when using Define a two-hour event-time watermark before stateful window aggregation. The scenario instead tests “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance”; the relevant result is The batch result has correct grouping without requiring a streaming checkpoint.

That observation is a valid indicator for “Events beyond the retention watermark should not cause unlimited state accumulation” when using Apply a suitable watermark and check the state cleanup behavior. The scenario instead tests “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance”; the relevant result is The batch result has correct grouping without requiring a streaming checkpoint.

Apply a suitable watermark and check the state cleanup behavior is appropriate when “Events beyond the retention watermark should not cause unlimited state accumulation”; it would be verified by Expired state stops growing with ancient event timestamps. It does not directly satisfy “A stateful stream joins two event streams on matching keys and bounded time difference,” whose relevant acceptance evidence is Late-data behavior and join state size match the bounded join window.

This action solves a different issue: “Late events within a two-hour tolerance must update event-time window aggregates.” Its normal signal is An event arriving within the watermark horizon updates the appropriate open window. Here the requested behavior is “A stateful stream joins two event streams on matching keys and bounded time difference,” so that signal is not sufficient.

Use ordinary batch aggregation instead of relying on streaming watermark state pruning belongs to the separate case “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance,” where The batch result has correct grouping without requiring a streaming checkpoint would be expected. That differs from the condition in this stem: “A stateful stream joins two event streams on matching keys and bounded time difference.”

This signal belongs to the adjacent case “Events beyond the retention watermark should not cause unlimited state accumulation,” whose remedy is Apply a suitable watermark and check the state cleanup behavior. The observed result for the present requirement is Late-data behavior and join state size match the bounded join window.

That observation is a valid indicator for “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance” when using Use ordinary batch aggregation instead of relying on streaming watermark state pruning. The scenario instead tests “A stateful stream joins two event streams on matching keys and bounded time difference”; the relevant result is Late-data behavior and join state size match the bounded join window.

The observed drop metrics correlate with event timestamps older than the watermark would demonstrate the separate action Inspect streaming progress and state operator late-event metrics alongside watermark settings, which addresses “Operators need to diagnose why late events were discarded from a windowed query.” It does not verify the requirement “A stateful stream joins two event streams on matching keys and bounded time difference” following Use a supported stream-stream join with event-time conditions and watermarks.

This action solves a different issue: “Events beyond the retention watermark should not cause unlimited state accumulation.” Its normal signal is Expired state stops growing with ancient event timestamps. Here the requested behavior is “Operators need to diagnose why late events were discarded from a windowed query,” so that signal is not sufficient.

Use ordinary batch aggregation instead of relying on streaming watermark state pruning is appropriate when “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance”; it would be verified by The batch result has correct grouping without requiring a streaming checkpoint. It does not directly satisfy “Operators need to diagnose why late events were discarded from a windowed query,” whose relevant acceptance evidence is The observed drop metrics correlate with event timestamps older than the watermark.

Use a supported stream-stream join with event-time conditions and watermarks belongs to the separate case “A stateful stream joins two event streams on matching keys and bounded time difference,” where Late-data behavior and join state size match the bounded join window would be expected. That differs from the condition in this stem: “Operators need to diagnose why late events were discarded from a windowed query.”

This signal belongs to the adjacent case “A batch query has no streaming state and the engineer mistakenly adds a watermark for performance,” whose remedy is Use ordinary batch aggregation instead of relying on streaming watermark state pruning. The observed result for the present requirement is The observed drop metrics correlate with event timestamps older than the watermark.

An event arriving within the watermark horizon updates the appropriate open window would demonstrate the separate action Define a two-hour event-time watermark before stateful window aggregation, which addresses “Late events within a two-hour tolerance must update event-time window aggregates.” It does not verify the requirement “Operators need to diagnose why late events were discarded from a windowed query” following Inspect streaming progress and state operator late-event metrics alongside watermark settings.

Late-data behavior and join state size match the bounded join window would demonstrate the separate action Use a supported stream-stream join with event-time conditions and watermarks, which addresses “A stateful stream joins two event streams on matching keys and bounded time difference.” It does not verify the requirement “Operators need to diagnose why late events were discarded from a windowed query” following Inspect streaming progress and state operator late-event metrics alongside watermark settings.