Databricks-Certified-Professional-Data-Engineer Databricks Certified Data Engineer Professional Exam Questions and Answers

Databricks provides different permission levels to control access to clusters. The correct minimal permission required for viewing driver logs is CAN VIEW.

Databricks Cluster Permission Levels:

CAN ATTACH TO:

Allows users to attach notebooks to a cluster but does not allow them to view logs.

Not sufficient for viewing driver logs.

CAN MANAGE:

Grants full control over the cluster, including starting, stopping, and editing configurations.

Too broad for this requirement.

CAN VIEW (Correct Answer):

Allows users to view cluster details, logs, and status but not modify any configurations.

Minimal required permission for viewing logs.

CAN RESTART:

Grants permission to restart the cluster, but does not include log access.

Not sufficient for viewing logs.

Conclusion:

The minimal permission needed to allow the development team to view driver logs is CAN VIEW.

https://www.databricks.com/blog/2020/08/31/introducing-the-databricks-web-terminal.html

The code is using %sh to execute shell code on the driver node. This means that the code is not taking advantage of the worker nodes or Databricks optimized Spark. This is why the code is taking longer to execute. A better approach would be to use Databricks libraries and APIs to read and write data from Git and DBFS, and to leverage the parallelism and performance of Spark. For example, you can use the Databricks Connect feature to run your Python code on a remote Databricks cluster, or you can use the Spark Git Connector to read data from Git repositories as Spark DataFrames.

This is the correct answer because the code uses the dropDuplicates method to remove any duplicate records within each batch of data before writing to the orders table. However, this method does not check for duplicates across different batches or in the target table, so it is possible that newly written records may have duplicates already present in the target table. To avoid this, a better approach would be to use Delta Lake and perform an upsert operation using mergeInto. Verified References: [Databricks Certified Data Engineer Professional], under “Delta Lake” section; Databricks Documentation, under “DROP DUPLICATES” section.

Delta Lake provides built-in versioning and time travel capabilities, allowing users to query previous snapshots of a table. This feature is particularly useful for understanding changes between different versions of the table. In this scenario, where the table is overwritten nightly, you can use Delta Lake's time travel feature to execute a query comparing the latest version of the table (the current state) with its previous version. This approach effectively identifies the differences (such as new, updated, or deleted records) between the two versions. The other options do not provide a straightforward or efficient way to directly compare different versions of a Delta Lake table.

In Databricks, using the Secrets module allows for secure management of sensitive information such as database credentials. Granting 'Read' permissions on a secret key that maps to database credentials for a specific team ensures that only members of that team can access these credentials. This approach aligns with the principle of least privilege, granting users the minimum level of access required to perform their jobs, thus enhancing security.

When triggering a job run using the Databricks CLI, the run_id field in the response represents a globally unique identifier for that particular run of the job. This run_id is distinct from the job_id. While the job_id identifies the job definition and is constant across all runs of that job, the run_id is unique to each execution and is used to track and query the status of that specific job run within the Databricks environment. This distinction allows users to manage and reference individual executions of a job directly.

This is the correct answer because it indicates a bottleneck caused by code executing on the driver. A bottleneck is a situation where the performance or capacity of a system is limited by a single component or resource. A bottleneck can cause slow execution, high latency, or low throughput. A production cluster has 3 executor nodes and uses the same virtual machine type for the driver and executor. When evaluating the Ganglia Metrics for this cluster, one can look for indicators that show how the cluster resources are being utilized, such as CPU, memory, disk, or network. If the overall cluster CPU utilization is around 25%, it means that only one out of the four nodes (driver + 3 executors) is using its full CPU capacity, while the other three nodes are idle or underutilized. This suggests that the code executing on the driver is taking too long or consuming too much CPU resources, preventing the executors from receiving tasks or data to process. This can happen when the code has driver-side operations that are not parallelized or distributed, such as collecting large amounts of data to the driver, performing complex calculations on the driver, or using non-Spark libraries on the driver. Verified References: [Databricks Certified Data Engineer Professional], under “Spark Core” section; Databricks Documentation, under “View cluster status and event logs - Ganglia metrics” section; Databricks Documentation, under “Avoid collecting large RDDs” section.

In a Spark cluster, the driver node is responsible for managing the execution of the Spark application, including scheduling tasks, managing the execution plan, and interacting with the cluster manager. If the overall cluster CPU utilization is low (e.g., around 25%), it may indicate that the driver node is not utilizing the available resources effectively and might be a bottleneck.

https://docs.delta.io/2.0.0/table-properties.html

 Delta Lake automatically collects statistics on the first 32 columns of each table, which are leveraged in data skipping based on query filters1. Data skipping is a performance optimization technique that aims to avoid reading irrelevant data from the storage layer1. By collecting statistics such as min/max values, null counts, and bloom filters, Delta Lake can efficiently prune unnecessary files or partitions from the query plan1. This can significantly improve the query performance and reduce the I/O cost.

The other options are false because:

Parquet compresses data column by column, not row by row2. This allows for better compression ratios, especially for repeated or similar values within a column2.

Views in the Lakehouse do not maintain a valid cache of the most recent versions of source tables at all times3. Views are logical constructs that are defined by a SQL query on one or more base tables3. Views are not materialized by default, which means they do not store any data, but only the query definition3. Therefore, views always reflect the latest state of the source tables when queried3. However, views can be cached manually using the CACHE TABLE or CREATE TABLE AS SELECT commands.

Primary and foreign key constraints can not be leveraged to ensure duplicate values are never entered into a dimension table. Delta Lake does not support enforcing primary and foreign key constraints on tables. Constraints are logical rules that define the integrity and validity of the data in a table. Delta Lake relies on the application logic or the user to ensure the data quality and consistency.

Z-order can be applied to any values stored in Delta Lake tables, not only numeric values. Z-order is a technique to optimize the layout of the data files by sorting them on one or more columns. Z-order can improve the query performance by clustering related values together and enabling more efficient data skipping. Z-order can be applied to any column that has a defined ordering, such as numeric, string, date, or boolean values.

Delta Lake's Liquid Clustering is an advanced feature that improves query performance by dynamically clustering data without requiring costly compaction steps like traditional Z-ordering.

When performing writes to a Liquid Clustered table, some write operations automatically maintain clustering, while others do not.

Explanation of Each Option:

(A) spark.writestream.format('delta').mode('append') (Correct Answer)

Reason: Streaming writes (writestream) do not support Liquid Clustering because streaming data arrives in micro-batches.

Since Liquid Clustering needs efficient global reorganization of files, streaming append operations don't provide sufficient data volume at a time to be effectively clustered.

Delta Lake documentation states that Liquid Clustering is only supported for batch writes.

(B) CTAS and RTAS statements

Reason: CREATE TABLE AS SELECT (CTAS) and REPLACE TABLE AS SELECT (RTAS) are batch operations and can enforce Liquid Clustering.

These operations create or replace a table based on a query result, and since they are batch-based, Liquid Clustering applies.

(C) INSERT INTO operations

Reason: INSERT INTO is supported for Liquid Clustering because it is a batch operation.

While it may not be as efficient as MERGE or COPY INTO, clustering is applied upon execution.

(D) spark.write.format('delta').mode('append')

Reason: Batch append operations are supported for Liquid Clustering.

Unlike streaming append, batch writes allow the optimizer to re-cluster data efficiently.

Conclusion:

Since streaming append operations do not support Liquid Clustering, option (A) is the correct answer.

Partitioning a Delta Lake table is a strategy used to improve query performance by dividing the table into distinct segments based on the values of a specific column. This approach allows queries to scan only the relevant partitions, thereby reducing the amount of data read and enhancing performance.

Considerations for Choosing a Partition Column:

Cardinality: Columns with high cardinality (i.e., a large number of unique values) are generally poor choices for partitioning. High cardinality can lead to a large number of small partitions, which can degrade performance.

Query Patterns: The partition column should align with common query filters. If queries frequently filter data based on a particular column, partitioning by that column can be beneficial.

Partition Size: Each partition should ideally contain at least 1 GB of data. This ensures that partitions are neither too small (leading to too many partitions) nor too large (negating the benefits of partitioning).

Evaluation of Columns:

date:

Cardinality: Typically low, especially if data spans over days, months, or years.

Query Patterns: Many analytical queries filter data based on date ranges.

Partition Size: Likely to meet the 1 GB threshold per partition, depending on data volume.

user_id:

Cardinality: High, as each user has a unique ID.

Query Patterns: While some queries might filter by user_id, the high cardinality makes it unsuitable for partitioning.

Partition Size: Partitions could be too small, leading to inefficiencies.

post_id:

Cardinality: Extremely high, with each post having a unique ID.

Query Patterns: Unlikely to be used for filtering large datasets.

Partition Size: Each partition would be very small, resulting in a large number of partitions.

post_time:

Cardinality: High, especially if it includes exact timestamps.

Query Patterns: Queries might filter by time, but the high cardinality poses challenges.

Partition Size: Similar to user_id, partitions could be too small.

Conclusion:

Given the considerations, the date column is the most suitable candidate for partitioning. It has low cardinality, aligns with common query patterns, and is likely to result in appropriately sized partitions.

The reason why the junior data engineer cannot transfer “Owner” privileges to the “DevOps” group is that Databricks jobs must have exactly one owner, and the owner must be an individual user, not a group. A job cannot have more than one owner, and a job cannot have a group as an owner. The owner of a job is the user who created the job, or the user who was assigned the ownership by another user. The owner of a job has the highest level of permission on the job, and can grant or revoke permissions to other users or groups. However, the owner cannot transfer the ownership to a group, only to another user. Therefore, the junior data engineer’s attempt to transfer “Owner” privileges to the “DevOps” group is not possible. References:

Jobs access control: https://docs.databricks.com/security/access-control/table-acls/index.html

Job permissions: https://docs.databricks.com/security/access-control/table-acls/privileges.html#job-permissions

In Databricks, catalogs, schemas (or databases), and tables are managed through the Unity Catalog or Hive Metastore, depending on the environment. Permissions and ownership within these structures are governed by access control lists (ACLs).

Catalog and Schema Ownership: When a platform engineer creates a catalog (such as catalog_A) and schema (such as schema_A), they automatically become the owner of those entities. This ownership gives them control over granting permissions for those entities (i.e., granting the USE CATALOG and USE SCHEMA privileges to others). However, ownership of the catalog or schema does not automatically extend to ownership or permission of individual tables within that schema.

Table Permissions: For tables within a schema, the permission model is more granular. The table creator (i.e., whoever creates the table) is automatically assigned as the owner of that table. In this case, the platform engineer owns the schema but does not automatically inherit permissions to any table created within the schema unless explicitly granted by the table's owner or unless they grant permissions to themselves.

Why the Engineer Lacks Access: The platform engineer notices that they do not have access to the underlying tables in schema_A despite being the owner of the schema. This occurs because the schema's ownership does not cascade to the tables. The engineer must either:

Grant permissions to themselves for the tables in schema_A, or

Be granted permissions by whoever created the tables within the schema.

Resolution: As the owner of the schema, the platform engineer can easily grant themselves the required permissions (such as SELECT, INSERT, etc.) for the tables in the schema. This explains why the owner of a schema may not automatically have access to the tables and must take explicit steps to acquire those permissions.

References

Databricks Unity Catalog Documentation: Manage Permissions

[Databricks Permissions and Ownership](https://docs.databricks.com/security/access-control/workspace-acl.html#permissions

Comprehensive and Detailed Explanation From Exact Extract:

Exact extract: “run_id: The canonical identifier of a run.”

This is the correct answer because it allows manual confirmation that these three requirements have been met. The requirements are that all tables containing Personal Identifiable Information (PII) must be clearly annotated, which includes adding column comments, table comments, and setting the custom table property “contains_pii” = true. The DESCRIBE EXTENDED command is used to display detailed information about a table, such as its schema, location, properties, and comments. By using this command on the dev.pii_test table, one can verify that the table has been created with the correct column comments, table comment, and custom table property as specified in the SQL DDL statement. Verified References: [Databricks Certified Data Engineer Professional], under “Lakehouse” section; Databricks Documentation, under “DESCRIBE EXTENDED” section.

The daily_store_sales table contains all the information needed to update store_sales_summary. The schema of the table is:

store_id INT, sales_date DATE, total_sales FLOAT

The daily_store_sales table is implemented as a Type 1 table, which means that old values are overwritten by new values and no history is maintained. The total_sales column might be adjusted after manual data auditing, which means that the data in the table may change over time.

The safest approach to generate accurate reports in the store_sales_summary table is to use Structured Streaming to subscribe to the change data feed for daily_store_sales and apply changes to the aggregates in the store_sales_summary table with each update. Structured Streaming is a scalable and fault-tolerant stream processing engine built on Spark SQL. Structured Streaming allows processing data streams as if they were tables or DataFrames, using familiar operations such as select, filter, groupBy, or join. Structured Streaming also supports output modes that specify how to write the results of a streaming query to a sink, such as append, update, or complete. Structured Streaming can handle both streaming and batch data sources in a unified manner.

The change data feed is a feature of Delta Lake that provides structured streaming sources that can subscribe to changes made to a Delta Lake table. The change data feed captures both data changes and schema changes as ordered events that can be processed by downstream applications or services. The change data feed can be configured with different options, such as starting from a specific version or timestamp, filtering by operation type or partition values, or excluding no-op changes.

By using Structured Streaming to subscribe to the change data feed for daily_store_sales, one can capture and process any changes made to the total_sales column due to manual data auditing. By applying these changes to the aggregates in the store_sales_summary table with each update, one can ensure that the reports are always consistent and accurate with the latest data. Verified References: [Databricks Certified Data Engineer Professional], under “Spark Core” section; Databricks Documentation, under “Structured Streaming” section; Databricks Documentation, under “Delta Change Data Feed” section.

This is the correct answer because it will allow the developer to update their local repository with the latest changes from the remote repository and switch to the desired branch. Pulling changes will not affect the current branch or create any conflicts, as it will only fetch the changes and not merge them. Selecting the dev-2.3.9 branch from the dropdown will checkout that branch and display its contents in the notebook. Verified References: [Databricks Certified Data Engineer Professional], under “Databricks Tooling” section; Databricks Documentation, under “Pull changes from a remote repository” section.

The scenario presented involves inconsistent microbatch processing times in a Structured Streaming job during peak hours, with the need to ensure that records are processed within 10 seconds. The trigger once option is the most suitable adjustment to address these challenges:

Understanding Triggering Options:

Fixed Interval Triggering (Current Setup): The current trigger interval of 10 seconds may contribute to the inconsistency during peak times as it doesn't adapt based on the processing time of the microbatches. If a batch takes longer to process, subsequent batches will start piling up, exacerbating the delays.

Trigger Once: This option allows the job to run a single microbatch for processing all available data and then stop. It is useful in scenarios where batch sizes are unpredictable and can vary significantly, which seems to be the case during peak hours in this scenario.

Implementation of Trigger Once:

Setup: Instead of continuously running, the job can be scheduled to run every 10 seconds using a Databricks job. This scheduling effectively acts as a custom trigger interval, ensuring that each execution cycle handles all available data up to that point without overlapping or queuing up additional executions.

Advantages: This approach allows for each batch to complete processing all available data before the next batch starts, ensuring consistency in handling data surges and preventing the system from being overwhelmed.

Rationale Against Other Options:

Option A and E (Decrease Interval): Decreasing the trigger interval to 5 seconds might exacerbate the problem by increasing the frequency of batch starts without ensuring the completion of previous batches, potentially leading to higher overhead and less efficient processing.

Option B (Increase Interval): Increasing the trigger interval to 30 seconds could lead to latency issues, as the data would be processed less frequently, which contradicts the requirement of processing records in less than 10 seconds.

Option C (Modify Partitions): While increasing parallelism through more shuffle partitions can improve performance, it does not address the fundamental issue of batch scheduling and could still lead to inconsistency during peak loads.

Conclusion:

By using the trigger once option and scheduling the job every 10 seconds, you ensure that each microbatch has sufficient time to process all available data thoroughly before the next cycle begins, aligning with the need to handle peak loads more predictably and efficiently.

References

Structured Streaming Programming Guide - Triggering

Databricks Jobs Scheduling

This is the correct answer because it exemplifies best practices for implementing this system. By isolating tables in separate databases based on data quality tiers, such as bronze, silver, and gold, the data engineering team can achieve several benefits. First, they can easily manage permissions for different users and groups through database ACLs, which allow granting or revoking access to databases, tables, or views. Second, they can physically separate the default storage locations for managed tables in each database, which can improve performance and reduce costs. Third, they can provide a clear and consistent naming convention for the tables in each database, which can improve discoverability and usability. Verified References: [Databricks Certified Data Engineer Professional], under “Lakehouse” section; Databricks Documentation, under “Database object privileges” section.

https://docs.databricks.com/en/delta/delta-change-data-feed.html

Regex, or regular expressions, are a powerful way of matching patterns in text. They can be used to identify key areas of text when parsing Spark Driver log4j output, such as the log level, the timestamp, the thread name, the class name, the method name, and the message. Regex can be applied in various languages and frameworks, such as Scala, Python, Java, Spark SQL, and Databricks notebooks. References:

https://docs.databricks.com/notebooks/notebooks-use.html#use-regular-expressions

https://docs.databricks.com/spark/latest/spark-sql/udf-scala.html#using-regular-expressions-in-udfs

https://docs.databricks.com/spark/latest/sparkr/functions/regexp_extract.html

https://docs.databricks.com/spark/latest/sparkr/functions/regexp_replace.html

Comprehensive and Detailed Explanation From Exact Extract:

Exact extract: “select() expects column names or Column expressions.”

Exact extract: “When using strings directly, Spark SQL interprets them as literal column names.”

Exact extract: “Python string operations, such as "colname"*3, return repeated strings, not column expressions.”

The expression 3*"heartrate" is Python string multiplication, which evaluates to "heartrateheartrateheartrate". The select() method interprets this as a literal column name. Since there is no column with that name in the DataFrame schema, Spark raises AnalysisException saying it cannot resolve that column. To correctly multiply a column by a scalar, one must use the column expression form:

from pyspark.sql.functions import col

df.select((col("heartrate") * 3).alias("heartrate_x3"))

This ensures Spark evaluates the arithmetic operation on the column instead of misinterpreting the string.

This is the correct answer because it explains which of the following adjustments will get a more accurate measure of how code is likely to perform in production. The adjustment is that calling display() forces a job to trigger, while many transformations will only add to the logical query plan; because of caching, repeated execution of the same logic does not provide meaningful results. When developing code in Databricks notebooks, one should be aware of how Spark handles transformations and actions. Transformations are operations that create a new DataFrame or Dataset from an existing one, such as filter, select, or join. Actions are operations that trigger a computation on a DataFrame or Dataset and return a result to the driver program or write it to storage, such as count, show, or save. Calling display() on a DataFrame or Dataset is also an action that triggers a computation and displays the result in a notebook cell. Spark uses lazy evaluation for transformations, which means that they are not executed until an action is called. Spark also uses caching to store intermediate results in memory or disk for faster access in subsequent actions. Therefore, calling display() forces a job to trigger, while many transformations will only add to the logical query plan; because of caching, repeated execution of the same logic does not provide meaningful results. To get a more accurate measure of how code is likely to perform in production, one should avoid calling display() too often or clear the cache before running each cell. Verified References: [Databricks Certified Data Engineer Professional], under “Spark Core” section; Databricks Documentation, under “Lazy evaluation” section; Databricks Documentation, under “Caching” section.

 This is the correct answer because Cmd 1 is written in Python and uses a list comprehension to extract the country names from the geo_lookup table and store them in a Python variable named countries af. This variable will contain a list of strings, not a PySpark DataFrame or a SQL view. Cmd 2 is written in SQL and tries to create a view named sales af by selecting from the sales table where city is in countries af. However, this command will fail because countries af is not a valid SQL entity and cannot be used in a SQL query. To fix this, a better approach would be to use spark.sql() to execute a SQL query in Python and pass the countries af variable as a parameter. Verified References: [Databricks Certified Data Engineer Professional], under “Language Interoperability” section; Databricks Documentation, under “Mix languages” section.

When joining a stream of advertisement impressions with a stream of user clicks, you want to minimize the state that you need to maintain for the join. Option A suggests using a left outer join with the condition that clickTime == impressionTime, which is suitable for correlating events that occur at the exact same time. However, in a real-world scenario, you would likely need some leeway to account for the delay between an impression and a possible click. It's important to design the join condition and the window of time considered to optimize performance while still capturing the relevant user interactions. In this case, having the watermark can help with state management and avoid state growing unbounded by discarding old state data that's unlikely to match with new data.

The code uses the DELETE FROM command to delete records from the users table that match a condition based on a join with another table called delete_requests, which contains all users that have requested deletion. The DELETE FROM command deletes records from a Delta Lake table by creating a new version of the table that does not contain the deleted records. However, this does not guarantee that the records to be deleted are no longer accessible, because Delta Lake supports time travel, which allows querying previous versions of the table using a timestamp or version number. Therefore, files containing deleted records may still be accessible with time travel until a vacuum command is used to remove invalidated data files from physical storage. Verified References: [Databricks Certified Data Engineer Professional], under “Delta Lake” section; Databricks Documentation, under “Delete from a table” section; Databricks Documentation, under “Remove files no longer referenced by a Delta table” section.

This is the correct answer because it accurately presents information about Delta Lake and Databricks that may impact the decision-making process of a junior data engineer who is trying to determine the best approach for dealing with schema declaration given the highly-nested structure of the data and the numerous fields. Delta Lake and Databricks support schema inference and evolution, which means that they can automatically infer the schema of a table from the source data and allow adding new columns or changing column types without affecting existing queries or pipelines. However, schema inference and evolution may not always be desirable or reliable, especially when dealing with complex or nested data structures or when enforcing data quality and consistency across different systems. Therefore, setting types manually can provide greater assurance of data quality enforcement and avoid potential errors or conflicts due to incompatible or unexpected data types. Verified References: [Databricks Certified Data Engineer Professional], under “Delta Lake” section; Databricks Documentation, under “Schema inference and partition of streaming DataFrames/Datasets” section.

DBFS is a file system protocol that allows users to interact with files stored in object storage using syntax and guarantees similar to Unix file systems1. DBFS is not a physical file system, but a layer over the object storage that provides a unified view of data across different data sources1. By default, the DBFS root is accessible to all users in the workspace, and the access to mounted data sources depends on the permissions of the storage account or container2. Mounted storage volumes do not need to have full public read and write permissions, but they do require a valid connection string or access key to be provided when mounting3. Both the DBFS root and mounted storage can be accessed when using %sh in a Databricks notebook, as long as the cluster has FUSE enabled4. The DBFS root does not store files in ephemeral block volumes attached to the driver, but in the object storage associated with the workspace1. Mounted directories will persist saved data to external storage between sessions, unless they are unmounted or deleted3. References: DBFS, Work with files on Azure Databricks, Mounting cloud object storage on Azure Databricks, Access DBFS with FUSE

In Databricks Lakehouse, to retain manually deleted or updated records in the raw_iot table while recomputing downstream tables when a pipeline update is run, the property pipelines.reset.allowed should be set to false. This property prevents the system from resetting the state of the table, which includes the removal of the history of changes, during a pipeline update. By keeping this property as false, any changes to the raw_iot table, including manual deletes or updates, are retained, and recomputation of downstream tables, such as bpm_stats, can occur with the full history of data changes intact.

Reading table’s changes, captured by CDF, using spark.read means that you are reading them as a static source. So, each time you run the query, all table’s changes (starting from the specified startingVersion) will be read.

This is the correct answer because it describes the behavior of Delta Lake Auto Compaction, which is a feature that automatically optimizes the layout of Delta Lake tables by coalescing small files into larger ones. Auto Compaction runs as an asynchronous job after a write to a table has succeeded and checks if files within a partition can be further compacted. If yes, it runs an optimize job with a default target file size of 128 MB. Auto Compaction only compacts files that have not been compacted previously. Verified References: [Databricks Certified Data Engineer Professional], under “Delta Lake” section; Databricks Documentation, under “Auto Compaction for Delta Lake on Databricks” section.

"Auto compaction occurs after a write to a table has succeeded and runs synchronously on the cluster that has performed the write. Auto compaction only compacts files that haven’t been compacted previously."

https://learn.microsoft.com/en-us/azure/databricks/delta/tune-file-size

This is the correct answer because the window function is used to group streaming data by time intervals. The window function takes two arguments: a time column and a window duration. The window duration specifies how long each window is, and must be a multiple of 1 second. In this case, the window duration is “5 minutes”, which means each window will cover a non-overlapping five-minute interval. The window function also returns a struct column with two fields: start and end, which represent the start and end time of each window. The alias function is used to rename the struct column as “time”. Verified References: [Databricks Certified Data Engineer Professional], under “Structured Streaming” section; Databricks Documentation, under “WINDOW” section. https://www.databricks.com/blog/2017/05/08/event-time-aggregation-watermarking-apache-sparks-structured-streaming.html

This is the correct answer because checkpointing is a critical feature of Structured Streaming that provides fault tolerance and recovery in case of failures. Checkpointing stores the current state and progress of a streaming query in a reliable storage system, such as DBFS or S3. Each streaming query must have its own checkpoint directory that is unique and exclusive to that query. If two streaming queries share the same checkpoint directory, they will interfere with each other and cause unexpected errors or data loss. Verified References: [Databricks Certified Data Engineer Professional], under “Structured Streaming” section; Databricks Documentation, under “Checkpointing” section.