Exam Bundle

That observation is a valid indicator for “A consumer is far behind and its required CDF versions were removed by retention” when using Perform a controlled full resync and establish a fresh version checkpoint. The scenario instead tests “A CDF pipeline applies duplicate changed rows under retry”; the relevant result is A replay leaves the downstream target logically unchanged.

Target state reflects the updated row once without double-counting would demonstrate the separate action Filter or reconcile change types to apply only intended current values, which addresses “A CDF update includes both preimage and postimage records.” It does not verify the requirement “A CDF pipeline applies duplicate changed rows under retry” following Deduplicate/merge idempotently by stable key and sequence.

This signal belongs to the adjacent case “Several independent downstream consumers need different checkpoints,” whose remedy is Store a durable version watermark per consumer. The observed result for the present requirement is A replay leaves the downstream target logically unchanged.

This action solves a different issue: “A new liquid clustered table is created but no clustering keys were selected automatically.” Its normal signal is The optimizer selects keys only when sufficient benefit is predicted. Here the requested behavior is “Queries commonly filter on different high-cardinality attributes that change over time,” so that signal is not sufficient.

The alternative Use liquid clustering or suitable layout instead of one partition per key has a distinct purpose—A high-cardinality filter benefits from clustering rather than massive Hive partitions. It would not establish the required outcome (Queries commonly filter on different high-cardinality attributes that change over time); the necessary verification concerns Filtering improves without exploding partition directory counts.

The alternative Remove pathological partitioning and redesign data layout has a distinct purpose—A daily writer creates millions of tiny partitions from a high-cardinality identifier. It would not establish the required outcome (Queries commonly filter on different high-cardinality attributes that change over time); the necessary verification concerns Filtering improves without exploding partition directory counts.

The table metadata and scan plan reflect its new clustering layout pertains to a different control (Stop ZORDER on the liquid-clustered table and run supported clustering maintenance) and a different need (An old Delta table uses ZORDER but is migrated to `CLUSTER BY`). It cannot establish that Use supported liquid clustering rather than creating thousands of tiny partitions met the stated goal.

File counts drop and partition scans no longer explode pertains to a different control (Remove pathological partitioning and redesign data layout) and a different need (A daily writer creates millions of tiny partitions from a high-cardinality identifier). It cannot establish that Use supported liquid clustering rather than creating thousands of tiny partitions met the stated goal.

File count and selective scan cost decrease without partition explosion pertains to a different control (Use liquid clustering or suitable layout instead of one partition per key) and a different need (A high-cardinality filter benefits from clustering rather than massive Hive partitions). It cannot establish that Use supported liquid clustering rather than creating thousands of tiny partitions met the stated goal.

Use liquid clustering or suitable layout instead of one partition per key is appropriate when “A high-cardinality filter benefits from clustering rather than massive Hive partitions”; it would be verified by File count and selective scan cost decrease without partition explosion. It does not directly satisfy “A relatively stable very low-cardinality partition key is required by legacy consumers,” whose relevant acceptance evidence is Files are grouped by the intended partition values without excessive fragmentation.

This action solves a different issue: “A CDF pipeline applies duplicate changed rows under retry.” Its normal signal is A replay leaves the downstream target logically unchanged. Here the requested behavior is “A relatively stable very low-cardinality partition key is required by legacy consumers,” so that signal is not sufficient.

This action solves a different issue: “A new liquid clustered table is created but no clustering keys were selected automatically.” Its normal signal is The optimizer selects keys only when sufficient benefit is predicted. Here the requested behavior is “A relatively stable very low-cardinality partition key is required by legacy consumers,” so that signal is not sufficient.

This signal belongs to the adjacent case “A large table is already liquid clustered on a new key but old files are not reorganized,” whose remedy is Run supported `OPTIMIZE` or automatic clustering maintenance. The observed result for the present requirement is Files are grouped by the intended partition values without excessive fragmentation.

Queries use enough scan tasks without excessive file-open overhead pertains to a different control (Choose a balanced target file size and tune supported write optimization) and a different need (Very large files cause poor parallelism for typical query scans). It cannot establish that Use a supported partition strategy with a small manageable partition count met the stated goal.

The table metadata and scan plan reflect its new clustering layout pertains to a different control (Stop ZORDER on the liquid-clustered table and run supported clustering maintenance) and a different need (An old Delta table uses ZORDER but is migrated to `CLUSTER BY`). It cannot establish that Use a supported partition strategy with a small manageable partition count met the stated goal.

Run supported `OPTIMIZE` or automatic clustering maintenance belongs to the separate case “A large table is already liquid clustered on a new key but old files are not reorganized,” where File layout changes and targeted query scan cost decreases would be expected. That differs from the condition in this stem: “A new liquid clustered table is created but no clustering keys were selected automatically.”

Use supported liquid clustering rather than creating thousands of tiny partitions is appropriate when “Queries commonly filter on different high-cardinality attributes that change over time”; it would be verified by Filtering improves without exploding partition directory counts. It does not directly satisfy “A new liquid clustered table is created but no clustering keys were selected automatically,” whose relevant acceptance evidence is The optimizer selects keys only when sufficient benefit is predicted.

The alternative Enable appropriate predictive optimization features for supported managed tables has a distinct purpose—A managed table's storage optimization should reflect observed query patterns automatically. It would not establish the required outcome (A new liquid clustered table is created but no clustering keys were selected automatically); the necessary verification concerns The optimizer selects keys only when sufficient benefit is predicted.

This signal belongs to the adjacent case “A team enables acceleration but sees no benefit because source data scanning dominates,” whose remedy is Inspect scan and I/O metrics before adjusting compute again. The observed result for the present requirement is The optimizer selects keys only when sufficient benefit is predicted.

Files are grouped by the intended partition values without excessive fragmentation pertains to a different control (Use a supported partition strategy with a small manageable partition count) and a different need (A relatively stable very low-cardinality partition key is required by legacy consumers). It cannot establish that Review predictive optimization eligibility and query workload instead of assuming an error met the stated goal.

Optimizer avoids unnecessary rewrites when predicted benefit is negligible would demonstrate the separate action Accept cost-aware optimizer selection rather than force arbitrary keys, which addresses “A very small table receives no automatic clustering keys despite configuration.” It does not verify the requirement “A new liquid clustered table is created but no clustering keys were selected automatically” following Review predictive optimization eligibility and query workload instead of assuming an error.

The alternative Use a supported partition strategy with a small manageable partition count has a distinct purpose—A relatively stable very low-cardinality partition key is required by legacy consumers. It would not establish the required outcome (An old Delta table uses ZORDER but is migrated to `CLUSTER BY`); the necessary verification concerns The table metadata and scan plan reflect its new clustering layout.

The alternative Run supported `OPTIMIZE` or automatic clustering maintenance has a distinct purpose—A large table is already liquid clustered on a new key but old files are not reorganized. It would not establish the required outcome (An old Delta table uses ZORDER but is migrated to `CLUSTER BY`); the necessary verification concerns The table metadata and scan plan reflect its new clustering layout.

Consume Change Data Feed instead of reprocessing the complete source table is appropriate when “Only a fraction of an upstream Delta table changes each hour”; it would be verified by The incremental job reads only applicable changes since checkpoint. It does not directly satisfy “An old Delta table uses ZORDER but is migrated to `CLUSTER BY`,” whose relevant acceptance evidence is The table metadata and scan plan reflect its new clustering layout.

File layout changes and targeted query scan cost decreases pertains to a different control (Run supported `OPTIMIZE` or automatic clustering maintenance) and a different need (A large table is already liquid clustered on a new key but old files are not reorganized). It cannot establish that Stop ZORDER on the liquid-clustered table and run supported clustering maintenance met the stated goal.

That observation is a valid indicator for “The planner initially picks a shuffle join but one input becomes broadcast-eligible after filtering” when using Allow AQE to adapt supported join strategy at runtime. The scenario instead tests “An old Delta table uses ZORDER but is migrated to `CLUSTER BY`”; the relevant result is The table metadata and scan plan reflect its new clustering layout.

Filtering improves without exploding partition directory counts would demonstrate the separate action Use supported liquid clustering rather than creating thousands of tiny partitions, which addresses “Queries commonly filter on different high-cardinality attributes that change over time.” It does not verify the requirement “An old Delta table uses ZORDER but is migrated to `CLUSTER BY`” following Stop ZORDER on the liquid-clustered table and run supported clustering maintenance.

Review predictive optimization eligibility and query workload instead of assuming an error is appropriate when “A new liquid clustered table is created but no clustering keys were selected automatically”; it would be verified by The optimizer selects keys only when sufficient benefit is predicted. It does not directly satisfy “A data engineer proposes partitioning by a unique transaction identifier,” whose relevant acceptance evidence is The table avoids one-file-per-key patterns and maintains manageable file count.

This action solves a different issue: “A daily writer creates millions of tiny partitions from a high-cardinality identifier.” Its normal signal is File counts drop and partition scans no longer explode. Here the requested behavior is “A data engineer proposes partitioning by a unique transaction identifier,” so that signal is not sufficient.

Use liquid clustering or suitable layout instead of one partition per key belongs to the separate case “A high-cardinality filter benefits from clustering rather than massive Hive partitions,” where File count and selective scan cost decrease without partition explosion would be expected. That differs from the condition in this stem: “A data engineer proposes partitioning by a unique transaction identifier.”

File layout changes and targeted query scan cost decreases would demonstrate the separate action Run supported `OPTIMIZE` or automatic clustering maintenance, which addresses “A large table is already liquid clustered on a new key but old files are not reorganized.” It does not verify the requirement “A data engineer proposes partitioning by a unique transaction identifier” following Reject the high-cardinality partition plan and choose clustering or a better low-cardinality layout.

This signal belongs to the adjacent case “Queries commonly filter on different high-cardinality attributes that change over time,” whose remedy is Use supported liquid clustering rather than creating thousands of tiny partitions. The observed result for the present requirement is The table avoids one-file-per-key patterns and maintains manageable file count.

File count and average file size improve in that table pertains to a different control (Execute a scoped OPTIMIZE maintenance job during a suitable window) and a different need (One unusually fragmented table needs a controlled immediate compaction). It cannot establish that Reject the high-cardinality partition plan and choose clustering or a better low-cardinality layout met the stated goal.