1

Resposta de referência

Schema evolution is managed with Schema Registry (Avro, Protobuf). Backward compatibility rules allow adding optional fields while avoiding breaking existing consumers.

2

Resposta de referência

- Driver Program - Initiates Spark application, and defines execution plan. - SparkContext - Coordinates tasks, manages resources, communicates with Cluster Manager. - Cluster Manager - Allocates resources, manages nodes in the Spark cluster. - Executor - Worker processes on cluster nodes, execute tasks, store data. - Task - Unit of work sent to Executor for execution. - RDD (Resilient Distributed Dataset) - Immutable, distributed collection of objects processed in parallel. - Spark Core - Foundation providing task scheduling, memory management, fault recovery. Also have Spark SQL, Spark Streaming, MLlib, GraphX, SparkR libs

3

Resposta de referência

Use GROUP BY on the columns that define uniqueness, with HAVING COUNT(*) > 1. Alternatively, use a window function like ROW_NUMBER() partitioned by key columns to identify and retrieve duplicates.

4

Resposta de referência

Hadoop is an open-source and scalable software framework used for distributed storage and processing of large amounts of data. Some of the reasons why Hadoop is used in business implementations are its features like: - Scalability - Flexibility - Easy to use and implement - Data Reliability and security - High level of fault tolerance

5

Resposta de referência

Spark is a MapReduce improvement in Hadoop. The difference between Spark and MapReduce is that Spark processes and retains data in memory for later steps, whereas MapReduce processes data on the disc. As a result, Spark's data processing speed is up to 100 times quicker than MapReduce for lesser workloads. Spark also constructs a Directed Acyclic Graph (DAG) to schedule tasks and orchestrate nodes throughout the Hadoop cluster, as opposed to MapReduce's two-stage execution procedure.

6

Resposta de referência

It is a column (like date) used to organize files into folders. This enables "Partition Pruning," allowing a query to skip entire folders of irrelevant data.

7

Resposta de referência

Type 0 retains the original value. Type 1 overwrites old data (no history). Type 2 adds a new row with a version flag or date range (full history). Type 3 adds a "previous value" column to the existing row.

8

Resposta de referência

New data applications are high-priced, so introducing such within a company doesn't happen that often. Nevertheless, when a company decides to invest in new data analytics tools, this could turn into quite an ambitious project. The new tools must be connected to the current systems in the company, and the employers who are going to use them should be formally trained. Additionally, maintenance of the tools should be administered and carried out on a regular basis. So, if you have prior experience, point out the obstacles you've overcome or list some scenarios of what could have gone wrong. In case you lack relevant experience, describe what you know about the process in detail. This will let the hiring manager know that, if a problem arises, you have the basic know-how that would help you through. Answer Example "As a data engineer, I've taken part in the introduction of a brand-new data analytics application in the last company I've worked for. The whole process requires a well-thought-out plan to ensure the smoothest transition possible. However, even the most careful planning can't rule out unforeseen issues. One of them was the high demand for user licenses which went beyond our expectations. The company had to reallocate financial resources to obtain additional licenses. Furthermore, training schedules had to be set up in a way that doesn't interrupt the workflow in different departments. In addition, we had to optimize our infrastructure, so that it could support the considerably higher number of users."

9

Resposta de referência

Synapse is a data warehouse service focused on querying structured data. Databricks is a unified analytics platform for big data and machine learning, with strong Spark-based processing.

10

Resposta de referência

-- DELETE: Removes specific rows, can be rolled back, logs each row DELETE FROM orders WHERE order_date < '2020-01-01'; -- TRUNCATE: Removes ALL rows, faster, minimal logging, resets identity TRUNCATE TABLE temp_staging; -- DROP: Removes the entire table structure DROP TABLE old_backup_table; | Command | Removes | Rollback? | Speed | Use Case | |---|---|---|---|---| | DELETE | Specific rows | Yes | Slow | Selective removal | | TRUNCATE | All rows | Limited | Fast | Clear staging tables | | DROP | Entire table | No | Fast | Remove unused tables | Important note: Exact behavior varies by database (e.g., transaction support, identity/sequence handling, logging). Why interviewers ask this: Running the wrong command in production is a classic mistake. Interviewers want to know you understand the consequences before touching production data.

11

Resposta de referência

Hadoop is suitable for long-running, batch-oriented jobs and when cost-effective storage is critical. Spark is more efficient for iterative and real-time workloads due to its in-memory processing. Spark has largely replaced MapReduce for most modern workloads due to its speed and developer flexibility.

12

Resposta de referência

Discuss serverless options, scalability, and cost considerations.

13

Resposta de referência

Optimizing a SQL query involves several strategies aimed at reducing the time and resources required to execute the query. Some common techniques include: - Indexing: Creating indexes on columns that are frequently used in WHERE, JOIN, and ORDER BY clauses can significantly speed up query performance by reducing the amount of data the database needs to scan. However, over-indexing can lead to slower write operations, so it's important to index judiciously. - Query Refactoring: Simplifying complex queries, breaking them into smaller parts, or removing unnecessary subqueries can improve performance. For example, instead of using a correlated subquery, consider using a JOIN or a WITH clause (common table expression) for better performance. - Avoiding SELECT *: Instead of selecting all columns with SELECT *, it's more efficient to explicitly list only the columns needed. This reduces the amount of data retrieved and processed by the database. - Using EXPLAIN Plan: The EXPLAIN or EXPLAIN ANALYZE command can be used to understand how the database is executing a query. It provides a query plan that shows which indexes are being used, how joins are performed, and where potential bottlenecks are. Example: Consider a scenario where you have a sales table with millions of rows, and you need to retrieve the total sales for a specific product over the last year. -- Original query SELECT product_id, SUM(sales_amount) FROM sales WHERE product_id = 12345 AND sales_date BETWEEN '2023-01-01' AND '2023-12-31' GROUP BY product_id; Optimized Query: -- Add an index on the product_id and sales_date columns CREATE INDEX idx_product_date ON sales(product_id, sales_date); -- Use the optimized query SELECT product_id, SUM(sales_amount) FROM sales WHERE product_id = 12345 AND sales_date BETWEEN '2023-01-01' AND '2023-12-31' GROUP BY product_id; In this example, creating an index on product_id and sales_date allows the database to quickly locate relevant rows, leading to a significant performance boost.

14

Resposta de referência

The CASE statement allows you to implement conditional logic in SQL queries. It evaluates conditions sequentially and returns a value when a condition is met. Suppose you want to categorize students based on their grades.' SELECT student_name, grade, CASE WHEN grade >= 90 THEN 'A' WHEN grade >= 80 THEN 'B' WHEN grade >= 70 THEN 'C' ELSE 'D' END AS grade_category FROM students; WHEN grade >= 90 THEN 'A' : Assigns 'A' to grades 90 or above.ELSE 'D' : Assigns 'D' to grades below 70.

15

Resposta de referência

Data normalization is the process of organizing data to reduce redundancy and improve data integrity. It involves dividing a database into two or more tables and defining relationships between them. It is important because it minimizes duplicate data, avoids anomalies, and ensures efficient storage and consistency.

16

Resposta de referência

Operators perform tasks (e.g., PythonOperator, BashOperator), while Sensors wait for a condition (e.g., file arrival, partition ready) before allowing downstream tasks to run.

17

Resposta de referência

This question reveals if you understand data processing beyond surface-level usage. Spark uses a cluster of machines to process data in parallel, breaking tasks into smaller units called “jobs” and “tasks” that run across multiple executors. Data is kept in memory when possible to avoid slow disk reads. A strong answer shows you grasp Spark's lazy evaluation, DAG scheduling, and the role of the driver and workers.

18

Resposta de referência

A star schema has a central fact table linked directly to denormalized dimension tables, making it simpler and faster for queries. In contrast, a snowflake schema normalizes the dimensions into multiple related tables, which reduces redundancy but can slow performance. Star schemas are often used in BI tools for speed, while snowflake schemas offer better data integrity and storage efficiency.

19

Resposta de referência

Describe a situation where you challenged existing processes or technologies. Explain the old way, why it needed change, your proposed innovation, and the positive results.

20

Resposta de referência

If the fact arrives after its dimension rows exist, I just load it normally with the correct surrogate key lookup based on the event timestamp. If the dimension is not there yet, I will insert a placeholder row with an inferred_flag so the fact still loads, then update it when the real dimension arrives. For reprocessing, I design pipelines to be idempotent over a rolling window — typically 7 to 14 days — using merge operations rather than append-only loads so reruns do not duplicate.

21

Resposta de referência

Watermarking tracks event-time progress and signals when a window of events is complete. It balances accuracy with latency by deciding when to stop waiting for late data. Without watermarks, systems risk either discarding valid data or delaying results indefinitely.

22

Resposta de referência

Memory in Python exists in the following way: - The objects and data structures initialized in a Python program are present in a private heap, and programmers do not have permission to access the private heap space. - You can allocate heap space for Python objects using the Python memory manager. The core API of the memory manager gives the programmer access to some of the tools for coding purposes. - Python has a built-in garbage collector that recycles unused memory and frees up memory for heap space.

23

Resposta de referência

Two important messages that NameNode gets from DataNode are

24

Resposta de referência

Optimizing SQL queries for large datasets involves several techniques to reduce execution time and resource consumption. This includes proper indexing to speed up data retrieval, using joins efficiently by ensuring that join conditions are on indexed columns and avoiding sub-queries and correlated sub-queries that can be rewritten as joins for better performance. Additionally, I leverage the ‘EXPLAIN' plan to understand how SQL queries are executed, which helps identify and optimize potential bottlenecks. Partitioning large tables and implementing query caching where appropriate also contribute to significant performance improvements, especially in environments with heavy read operations.

25

Resposta de referência

An open-source platform for manipulating and storing data in Hadoop. Moreover, it is used for running applications in clusters. The primary benefit is the vast volumes of space required for data storage and a tremendous amount of processing power to manage an infinite number of jobs and tasks simultaneously. The three different modes of Hadoop are: - Standalone mode - Pseudo distributed mode - Fully distributed mode.

26

Resposta de referência

GDPR (General Data Protection Regulation) is a regulation in EU law on data protection and privacy. For data engineering, it impacts: - Data collection and storage practices - Data processing and usage - Data subject rights (e.g., right to be forgotten) - Data breach notification requirements - Cross-border data transfers

27

Resposta de referência

When asked about cost control, explain that you reduce expenses by choosing efficient file formats (Parquet, ORC), applying partitioning and clustering, caching intermediate results, and cleaning up unused data. Highlight monitoring and cost dashboards to track spend and optimize storage tiers. You should also mention tuning compute resources and autoscaling policies. This shows interviewers that you not only build pipelines but also keep an eye on business value.

28

Resposta de referência

Keep it professional; don't badmouth coworkers. Show you listened, understood their view, and found a solution that worked for both sides.

29

Resposta de referência

Event-driven architecture enables systems to respond instantly to events, ideal for real-time analytics, monitoring, and automation. Azure Event Hubs, a scalable event ingestion service, works seamlessly with Azure Functions for processing. Workflow with Event Hubs and Azure Functions: - Event producer: Apps, IoT devices, or logs send events to Event Hubs. - Processing: Azure Function listens to the stream and triggers custom logic. - Output: Processed events go to Cosmos DB, SQL, Data Lake, or Power BI.

30

Resposta de referência

Align your goals with growth at Amazon. For example: 'I aim to become a technical leader in data engineering, designing large-scale data systems and mentoring junior engineers.'

31

Resposta de referência

- Use event-driven architecture (e.g., Cloud Functions, Lambda triggers) - Decouple compute from storage (S3, GCS, ADLS) - Build idempotent, retry-safe ETL jobs - Use managed orchestration tools like Cloud Composer or Azure Data Factory

32

Resposta de referência

Provide an example where you used data, logical reasoning, and persuasive communication to shift the team's opinion. Explain the situation, your alternative proposal, and the successful outcome.

33

Resposta de referência

Identify which jobs are affected and isolate the failure. Restore critical workflows with a temporary fix. Add schema validation or contract checks to catch future changes early. Communicate with the upstream team to coordinate. Implement monitoring for schema shifts.

34

Resposta de referência

Since most companies are now shifting to cloud-based environments, this question lets the interviewer know how prepared you are to work in a cloud-based environment. You should show your preparedness and familiarity with the cloud-based environment along with the pros of cloud computing such as: - Its flexibility and scalability. - Security and mobility. - Risk-free data access from anywhere.

35

Resposta de referência

Cloud computing provides scalable storage and compute resources, managed databases, and services (like data warehousing, ETL tools), simplifying infrastructure management and accelerating data pipeline development.

36

Resposta de referência

A common example is building a clickstream pipeline. Kafka ingests user activity events, Flink or Spark Streaming processes and aggregates them, and results are stored in a warehouse or NoSQL database. Observability and exactly-once guarantees ensure reliability and correctness.

37

Resposta de referência

Azure Synapse is a cloud data warehouse designed for analytics and BI workloads. Azure Data Lake, on the other hand, stores raw structured and unstructured data at scale. Synapse is optimized for queries and reporting, while Data Lake serves as a foundation for transformations and ML pipelines.

38

Resposta de referência

ETL extracts data from sources, transforms it (cleans, formats), and loads it into a data warehouse or database. It's the core process for moving and preparing data.

39

Resposta de referência

First identify duplicates using GROUP BY and COUNT or window functions like ROW_NUMBER(). Then determine the deduplication rule based on business logic, such as keeping the latest record by timestamp. Finally, use DELETE or MERGE statements to remove duplicates while maintaining data integrity.

40

Resposta de referência

To identify duplicate rows, you can use the GROUP BY clause along with the HAVING clause to filter groups with more than one occurrence. SELECT column1, column2, ..., COUNT(*) AS duplicate_count FROM table_name GROUP BY column1, column2, ... HAVING COUNT(*) > 1; Suppose you have a table employees with columns id , name , and department . To find duplicate employees based on name and department : SELECT name, department, COUNT(*) AS duplicate_count FROM employees GROUP BY name, department HAVING COUNT(*) > 1; GROUP BY : Groups rows by the specified columns (name anddepartment ).COUNT(*) : Counts the number of rows in each group.HAVING COUNT(*) > 1 : Filters groups with more than one occurrence, indicating duplicates.

41

Resposta de referência

They allow you to cache a small, read-only variable on every worker node once, rather than sending it with every task. This is used in "Broadcast Joins" to avoid shuffling a small table.

42

Resposta de referência

In a relational database, data is organized into tables, and the way tables relate to each other establishes different types of relationships. - One-to-One (1:1): This is a rare relationship where each record in Table A links to one and only one record in Table B, and vice versa. Example: A table of students with one unique health record each. - One-to-Many (1:M): The most common relationship where a record in Table A can relate to one or several records in Table B, but any single record in Table B links back to only one record in Table A. Example: A customer can have multiple orders, but each order is associated with only one customer. - Many-to-Many (M:N): This type of relationship initially presents itself as 1:M on both sides. To handle M:N relationships, a special table, known as a "junction table" or "associate table," is introduced. This table typically consists of composite primary keys, one from each of the two related tables. Example: In a library database, a book can have multiple authors, and an author can write multiple books. The relationship between the "Book" and "Author" tables is M:N, so a junction table, say "Book_Author," is created.

43

Resposta de referência

Framework for answering: 1. Clarify requirements first: - Where does the source data come from? (Database? Files? API?) - How much data per day? (This affects tool choice) - Who consumes the output? (Analysts? Dashboards? ML models?) - What's the latency requirement? (By 6 AM? Within 1 hour of data arriving?) 2. Propose a high-level architecture: Source DB → [Extract] → Raw Storage → [Transform] → Data Warehouse → BI Tool (Python) (S3/GCS) (Spark/dbt) (Snowflake) (Tableau) 3. Address key concerns: - Scheduling: “I'd use Airflow to orchestrate, running at 2 AM after source systems close” - Error handling: “Add alerts on failure, implement retries with exponential backoff” - Data quality: “Run validation checks before loading to production tables” - Idempotency: “Use delete-insert pattern for daily partitions so reruns are safe” Why interviewers ask this: They want to see structured thinking, not perfect answers. Ask clarifying questions. State your assumptions. Explain tradeoffs.

44

Resposta de referência

The Combiner in Hadoop's MapReduce acts as a mini-reducer during the Map phase, processing outputs locally to minimize the data shuffled across the network, significantly boosting the efficiency of the MapReduce jobs. However, using a Combiner must be appropriate because it does not change the reducer's output. It should be used only when the operation is commutative and associative, such as summing numbers or finding a maximum.

45

Resposta de referência

In a distributed join (Shuffle), data is moved across the network to align keys. This is expensive. In a Broadcast Join, the smaller table is copied (broadcasted) to every worker node. The large table does not move. This eliminates the network shuffle and drastically improves performance for Large-to-Small table joins.

46

Resposta de referência

I use encryption at rest and in transit, implement data masking (hiding sensitive parts of a string), and use Role-Based Access Control (RBAC) to limit who can see the data.

47

Resposta de referência

Some of the must-have skills for data engineers that you should mention are: - Detailed understanding of data modeling - Knowledge of data warehousing tools like SQL and NoSQL - Data visualization and transformation knowledge - Experience with distributed systems like Hadoop, Spark, etc. - Knowledge of data warehousing and ETL tools - Ability to think out of the box and understand the requirement of the business team to convert raw data into a structured format - Robust mathematical, statistical, and computational skills - Programming knowledge in tools like Python, Java, Javascript, and others

48

Resposta de referência

Ensuring data quality involves: - Data Validation: Checking data against predefined rules. - Data Cleaning: Removing duplicates, handling missing values, and correcting errors. - Monitoring: Continuously tracking data quality metrics. - Automated Testing: Implementing tests to catch quality issues early in the pipeline.

49

Resposta de referência

This is a question that interviewers may ask to test your SQL expertise. To reduce duplicate data points, you can advise using the SQL keywords DISTINCT & UNIQUE. You should also provide additional approaches, such as utilizing GROUP BY to deal with duplicate data items.

50

Resposta de referência

HDFS stands for Hadoop Distributed File System and handles large data sets running on particular hardware. HDFS acts as the primary data storage option and employs the NameNode and DataNode architecture to enable users to retrieve and store information in a scalable Hadoop cluster easily.

51

Resposta de referência

Normalization is a set of techniques used in database design to ensure data integrity, reduce redundancy, and improve overall performance. It involves structuring data into multiple related tables, each serving a specific purpose. - Primary Keys: Unique identifiers for each record. - Foreign Keys: Links tables, establishing relationships. - Atomicity: Ensuring each data field is singular, not compound. - Data Consistency: Avoids contradictory or outdated information. - Minimizes Redundancy: Saves storage and prevents update anomalies. - Simplifies Queries: Easier to construct and understand. - First Normal Form (1NF): Data is atomic. - Second Normal Form (2NF): In 1NF, and all non-key attributes are fully functionally dependent on the primary key. - Third Normal Form (3NF): In 2NF, and there are no transitive dependencies. Beyond 3NF: - Boyce-Codd Normal Form (BCNF): A specialized version of 3NF where each determinant is a candidate key. - Fifth Normal Form (5NF): Achieved through decomposition to the point where no further decomposition is possible, ensuring "join dependency" consistency.

52

Resposta de referência

A: Key differences include: - Speed: Spark is generally faster due to in-memory processing - Ease of use: Spark offers more user-friendly APIs in multiple languages - Versatility: Spark supports various workloads beyond batch processing, including streaming and machine learning - Iterative processing: Spark is more efficient for iterative algorithms common in machine learning

53

Resposta de referência

OLTP (Online Transactional Processing) is optimized for frequent, small transactions and fast writes (e.g., banking). OLAP (Online Analytical Processing) is optimized for complex data analysis and heavy read operations over massive datasets (e.g., sales trends).

54

Resposta de referência

Optimizing a SQL query involves several strategies aimed at reducing the time and resources required to execute the query. Some common techniques include: - Indexing: Creating indexes on columns that are frequently used in WHERE, JOIN, and ORDER BY clauses can significantly speed up query performance by reducing the amount of data the database needs to scan. However, over-indexing can lead to slower write operations, so it's important to index judiciously. - Query Refactoring: Simplifying complex queries, breaking them into smaller parts, or removing unnecessary subqueries can improve performance. For example, instead of using a correlated subquery, consider using a JOIN or a WITH clause (common table expression) for better performance. - Avoiding SELECT *: Instead of selecting all columns with SELECT *, it's more efficient to explicitly list only the columns needed. This reduces the amount of data retrieved and processed by the database. - Using EXPLAIN Plan: The EXPLAIN or EXPLAIN ANALYZE command can be used to understand how the database is executing a query. It provides a query plan that shows which indexes are being used, how joins are performed, and where potential bottlenecks are. Example: Consider a scenario where you have a sales table with millions of rows, and you need to retrieve the total sales for a specific product over the last year. -- Original query SELECT product_id, SUM(sales_amount) FROM sales WHERE product_id = 12345 AND sales_date BETWEEN '2023-01-01' AND '2023-12-31' GROUP BY product_id; Optimized Query: -- Add an index on the product_id and sales_date columns CREATE INDEX idx_product_date ON sales(product_id, sales_date); -- Use the optimized query SELECT product_id, SUM(sales_amount) FROM sales WHERE product_id = 12345 AND sales_date BETWEEN '2023-01-01' AND '2023-12-31' GROUP BY product_id; In this example, creating an index on product_id and sales_date allows the database to quickly locate relevant rows, leading to a significant performance boost.

55

Resposta de referência

min(largest_values) finds the smallest element in largest_values. largest_values.index(min_val) gets the index of this smallest element so it can be replaced with a new larger element. - The final sorted (largest_values, reverse=True) call is optional, depending on whether you want the results sorted in descending order. def find_largest(input_list, m): # Check for edge cases if not input_list or m <= 0: return None # Initialize list to store the largest m values largest_values = [] for num in input_list: if len(largest_values) < m: # Add to the list if we haven't found m elements yet largest_values.append(num) else: # Find the smallest element in largest_values min_val = min(largest_values) if num > min_val: # Replace the smallest element if current num is larger min_index = largest_values.index(min_val) largest_values[min_index] = num # Optional: Sort in descending order return sorted(largest_values, reverse=True) # Example usage: input_list = [3, 1, 5, 6, 8, 2, 9, 10, 7] m = 3 print(find_largest(input_list, m)) # Output should be [10, 9, 8]

56

Resposta de referência

Tailor this answer to your experience. For example: "At my previous role, I designed a real-time fraud detection pipeline using Kafka for event ingestion, Spark Streaming for processing, and Elasticsearch for storing anomalies. We scaled to 500K messages/minute and implemented alerting using Grafana and Prometheus."

57

Resposta de referência

Macros in Excel refers to an action or a set of actions that can be saved and recorded to run as often as required. Macros may be given names and can be used to save time to perform any frequently run tasks. Excel stores macros as VBA code, and you can view the code using a VBA editor. You can assign macros to objects, including shapes, graphics, or control.

58

Resposta de referência

Azure Databricks and Azure Machine Learning (Azure ML) can be integrated for scalable data processing, model training, and deployment in the cloud. Integration steps: - Data prep in Databricks: Use Apache Spark to clean and transform large datasets. - Train and register model: Train models (e.g., MLlib, Scikit-learn) in Databricks and register them in Azure ML. - Deploy via Azure ML: Use Azure ML Managed Online Endpoints for real-time inference. - Automate with Pipelines: Schedule workflows using Azure ML Pipelines or Databricks Jobs.

59

Resposta de referência

Processing of data in large, distinct fragments is batch processing. In contrast, processing real-time data, one record at a time, is stream processing. When delay in processing is not a significant concern, we go for batch processing. e.g., generating daily reports or historical data analysis. When we cannot afford delay in processing, we use stream processing. e.g., real-time analytics, monitoring, and fraud detection.

60

Resposta de referência

A portmanteau of "Update" and "Insert." It checks if a record exists; if it does, it updates the record; if it doesn't, it inserts a new one. This is often handled by the MERGE statement in SQL.

61

Resposta de referência

Apache Spark is an open-source, distributed computing system providing fast, in-memory data processing for big data analytics. Spark is faster, more versatile, and developer-friendly compared to MapReduce, offering in-memory processing and a broader range of libraries for big data analytics. - Spark performs in-memory processing, reducing disk I/O and speeding up tasks. MapReduce reads and writes to disk, making it slower for iterative algorithms. - Spark offers high-level APIs in multiple languages, making development more accessible. MapReduce involves more complex and verbose code. - Spark is well-suited for iterative algorithms due to in-memory caching. MapReduce is less efficient for iterative tasks

62

Resposta de referência

Profile the data to understand patterns. Decide on handling strategy: impute, drop, or flag missing values. Standardize formats (dates, strings). Validate against business rules. Document all cleaning steps for reproducibility.

63

Resposta de referência

Data Profiling indicates analyzing and gathering information about data. Thus, it helps analyze the data quality and further data processing, like cleaning or transforming.

64

Resposta de referência

COSHH is an acronym for Classification and Optimization-based Scheduling for Heterogeneous Hadoop systems. As the name implies, it offers scheduling at both the cluster and application levels to speed up job completion.

65

Resposta de referência

With this question, the panel generally wants to know your problem-solving ability and how well you perform under pressure. To answer the question, first, brief them about the situations that lead to the problem. You should tell them about your role in that situation. For example, if you played a leading role in solving that problem, that would tell the interviewer about competency as a leader. After that tell them about the action you took to solve the problem. To end the answer on a positive note, you should tell them about the consequences of the challenge and the learning you took out of it.

66

Resposta de referência

Data modeling is critical in data engineering, offering a systematic framework for data storage, processing, and retrieval that supports effective data management and operational efficiency. It allows engineers to define the data's logical structure and establish relationships between models. This process is crucial for developing efficient databases and helps visualize complex data relationships, making it easier for stakeholders to understand the data architecture and make informed decisions. Data modeling enhances data quality and reduces redundancy, which is vital for any scalable data system by ensuring that all data interactions are logically planned.

67

Resposta de referência

Answer with a well-known example like Zappos or Amazon itself. Explain specific practices: 24/7 support, no-questions-asked returns, personalized interactions, and how they prioritize customer satisfaction over short-term profit.

68

Resposta de referência

Answer by walking through: - The business problem and scope of the project. - Technical challenges (e.g., data volume, integration complexity). - Team coordination and stakeholder management. - The solution you guided the team to implement. - The outcome and lessons learned.

69

Resposta de referência

For deduplication, use window functions in SQL or Pandas drop_duplicates() in Python.

70

Resposta de referência

When discussing SQL commands in ETL, focus on their roles: SELECT retrieves data, JOIN combines tables based on relationships, WHERE filters specific records, ORDER BY sorts results, and GROUP BY aggregates data for analysis. Emphasize understanding how to use these commands effectively to extract, transform, and load data, ensuring clarity in data manipulation and retrieval processes.

71

Resposta de referência

Describe a situation where you delivered more than expected. For example: 'I was asked to clean a dataset, but I also built a validation dashboard and documented the process, which the team used for future projects.'

72

Resposta de referência

dbt (data build tool) transforms data inside your warehouse using SQL. It's the “T” in ELT. Key features: - Write transformations as SQL SELECT statements - Automatic dependency management between models - Built-in testing and documentation - Version control friendly (SQL files in git) -- dbt model: models/marts/sales_summary.sql {{ config(materialized='table') }} SELECT date_trunc('month', order_date) as month, product_category, SUM(amount) as total_sales, COUNT(DISTINCT customer_id) as unique_customers FROM {{ ref('stg_orders') }} GROUP BY 1, 2 Why interviewers ask this: dbt has become standard for analytics engineering. Understanding it shows you're current with industry practices.

73

Resposta de referência

Documentation is stored in schema.yml files and compiled into a dbt docs site, showing lineage graphs, descriptions, and test coverage.

74

Resposta de referência

There are three modes in Hadoop, namely

75

Resposta de referência

We can address the eventual consistency by enforcing tools like Conflict Resolution Strategies (e.g., Last Write Wins), Version Vectors, or Quorum-based Replication to ensure that, over time, all duplicates combine to the same state.

76

Resposta de referência

Basically the question asks is to avoid full fledged ETL. You can use CDC, like Incremental Load for this month, day, or time window like filter down the source or we can use partitioning in Source based on day or month or watermarking is also the best way in data modelling to check for new ones.

77

Resposta de referência

A: Popular programming languages for data engineering include: - Python - SQL - Java - Scala - R

78

Resposta de referência

A self-join is when a table is joined with itself. It is commonly used for hierarchical data, such as a table where an employee_id is linked to a manager_id in the same table.

79

Resposta de referência

- AWS CloudTrail allows security analysis, resource change tracking, and compliance auditing of an AWS environment by providing a history of AWS API calls for an account. CloudTrail sends log files to a chosen Amazon Simple Storage Service (Amazon S3) bucket, with optional log file integrity validation. - Amazon S3 Access Logs record individual requests to Amazon S3 buckets and can be capable of monitoring traffic patterns, troubleshooting, and security and access audits. It can also assist a business in gaining a better understanding of its client base, establishing lifecycle policies, defining access policies, and determining Amazon S3 prices. - Amazon VPC Flow Logs record IP traffic between Amazon Virtual Private Cloud (Amazon VPC) network interfaces at the VPC, subnet, or individual Elastic Network Interface level. You can store Flow log data in Amazon CloudWatch Logs and export it to Amazon CloudWatch Streams for enhanced network traffic analytics and visualization.

80

Resposta de referência

Both of these are functions that data engineers should know. The *args function enables users to specify ordered functions to use in the command line. Meanwhile, the **kwarg function expresses a group of in-line and unordered arguments that must be passed to a function.

81

Resposta de referência

The reason why data engineers mostly work “backstage” is that making data available comes much earlier in the data analysis project timeline. That said, c-level executives in the company are usually more interested in the later stages of the work process. More specifically, their goal is to understand the insights that data scientists extract from the data via statistical and machine learning models. So, your answer to this question will tell the hiring manager if you're only able to work in the spotlight, or if you thrive in both situations. Answer Example "As a data engineer, I realize that I do most of my work away from the spotlight. But that has never been that important to me. I believe what matters is my expertise in the field and how it helps the company reach its goals. However, I'm pretty comfortable being in the spotlight whenever I need to be. For example, if there's a problem in my department which needs to be addressed by the company executives, I won't hesitate to bring their attention to it. I think that's how I can further improve my team's work and reach better results for the company."

82

Resposta de referência

Methods to stay updated include: - Following relevant blogs, podcasts, and YouTube channels - Participating in online communities (e.g., Stack Overflow, Reddit) - Attending webinars and virtual conferences - Subscribing to industry newsletters - Networking with other professionals in the field - Experimenting with new tools and technologies in personal projects

83

Resposta de referência

As a manager, decision-making involves balancing technical tradeoffs with business priorities. Strong leaders: - Prioritize work based on business impact and dependencies. - Gather input from engineers, stakeholders, and data consumers. - Foster a culture of experimentation and learning. - Communicate decisions clearly, including rationale and tradeoffs.

84

Resposta de referência

Schema-on-write enforces structure before data is stored (common in data warehouses). Schema-on-read applies structure when data is accessed (common in data lakes). Understanding this trade-off helps in choosing the right architecture.

85

Resposta de referência

Kafka achieves fault tolerance through replication. Each partition can have multiple replicas across different brokers. Data is persisted to disk and can be retained for a configurable period, ensuring durability even if consumers fail to consume it immediately.

86

Resposta de referência

I would identify bottlenecks through profiling. Strategies include using incremental extraction instead of full loads, parallelizing transformations, partitioning large tables, optimizing join logic, using columnar storage formats like Parquet, increasing cluster resources, and caching intermediate results.

87

Resposta de referência

A Data Warehouse is a centralized storage system designed for query and analysis, integrating structured data from multiple sources. For example, using Snowflake to store sales, marketing, and CRM data is a typical use case. A Data Lake, on the other hand, is a centralized repository for storing structured, semi-structured, and unstructured data at scale. It is more flexible and is often used to store raw data, such as IoT feeds and logs. For instance, Azure Data Lake can store diverse data types for future processing. Key Difference: Data Warehouses are optimized for analytics on structured data, while Data Lakes handle unstructured data with less rigid schema requirements.

88

Resposta de referência

Hadoop uses a batch processing model and stores data on disk between each operation, which makes it slower. Spark, on the other hand, processes data in-memory, offering much faster performance for iterative and real-time tasks. While Hadoop is suited for long-running jobs on massive datasets, Spark is preferred for complex analytics, machine learning, and streaming use cases. Spark also supports more user-friendly APIs in Python, Scala, and SQL.

89

Resposta de referência

Approaching data security in data engineering projects involves implementing a combination of best practices, tools, and policies to protect data at all stages of its lifecycle—during collection, storage, processing, and transmission. Key Strategies: - Data Encryption: - At Rest: Ensure that all sensitive data is encrypted at rest using strong encryption algorithms like AES-256. This applies to databases, data lakes, and any storage services used in the project. - In Transit: Data should also be encrypted in transit using protocols like TLS (Transport Layer Security) to protect it from interception during transmission between systems. - Access Control: - Implement strict access control mechanisms to ensure that only authorized users and systems can access the data. This involves using role-based access control (RBAC) and enforcing the principle of least privilege, where users are given the minimum access necessary to perform their tasks. - Use IAM (Identity and Access Management) tools provided by cloud platforms (e.g., AWS IAM, Google Cloud IAM) to manage and audit access permissions. - Data Masking and Anonymization: - For sensitive data, implement data masking or anonymization techniques to protect personally identifiable information (PII) while still allowing the data to be used for analysis. Techniques like tokenization or pseudonymization can be used to obscure sensitive details. - Audit Logging: - Maintain detailed audit logs of all data access and processing activities. These logs should capture who accessed the data, what actions were taken, and when they occurred. Audit logs are essential for detecting unauthorized access and for compliance with regulations like GDPR or HIPAA. - Regular Security Audits and Penetration Testing: - Conduct regular security audits and penetration testing to identify and address vulnerabilities in the data infrastructure. This includes reviewing configurations, patching software, and ensuring compliance with security policies. - Data Governance and Compliance: - Implement data governance policies to ensure that data is managed and protected according to legal and regulatory requirements. This includes defining data ownership, handling data classification, and ensuring compliance with data protection laws like GDPR, CCPA, or HIPAA.

90

Resposta de referência

- Azure has a separate analytics service called Azure Stream Analytics, which supports the Stream Analytics Query Language, a primary SQL-based language. - It enables you to extend the query language's capabilities by introducing new Machine Learning functions. - Azure Stream Analytics can analyze a massive volume of structured and unstructured data at around a million events per second and provide relatively low latency outputs.

91

Resposta de referência

The medallion architecture organizes data into bronze (raw), silver (cleaned), and gold (aggregated) layers. It breaks down at scale when data volumes are extremely high, requiring more granular partitioning or when the governance overhead of multiple layers becomes costly.

92

Resposta de referência

This question exposes whether you understand data tools at different scales. Pandas is perfect for small to medium-sized data on a single machine. PySpark is built for distributed computing and can handle massive datasets across clusters. Interviewers want to see if you know when to switch between them instead of forcing one tool to do everything.

93

Resposta de referência

Common bottlenecks include insufficient indexing, poorly written SQL (e.g., full scans, large joins), network latency, I/O limits on storage, memory constraints in processing engines, and contention in orchestration. Monitoring and profiling help identify them.

94

Resposta de referência

The Interviewer's Goal: To see if you wait for a CEO to find a bug, or if you catch it automatically. The Answer: Data quality isn't a one-time fix; it is a continuous process known as Data Observability. I implement automated checks at three specific stages: - Volume Checks: Detecting 'Silent Failures.' (e.g., 'We usually get 1 million rows/day. Today we got 500. Alert the team immediately.') - Schema Validation: Ensuring the source system didn't change a data type (e.g., changing a UserID from Integer to String). - Distribution/Statistical Checks: Detecting logic errors. (e.g., 'The average order value is usually $50. Today it is $5,000. Something is wrong.') I rely on tools like Great Expectations, dbt tests, or Soda to block bad data before it hits production dashboards.

95

Resposta de referência

A streaming-first engine that processes every event individually (true streaming), offering lower latency than Spark's micro-batching approach.

96

Resposta de referência

COSHH, which stands for Classification and Optimization based Schedule for Heterogeneous Hadoop systems, is essential for optimizing task scheduling and resource allocation within Hadoop clusters. It aims to improve the efficiency of running Hadoop jobs by classifying jobs based on resource requirements and the heterogeneity of available resources. By using COSHH, Hadoop can better manage resources across different nodes, significantly reducing the completion time of jobs and improving overall system performance. This aspect is especially crucial in environments characterized by clusters with nodes of varying capacities and workloads, necessitating adaptable and efficient resource management.

97

Resposta de referência

Blocks play a pivotal role in distributing data across the Hadoop cluster, breaking down large data sets into manageable pieces. This ensures efficient storage and processing. On the other hand, Block Scanners meticulously verify the integrity of the data stored in these blocks, safeguarding against corruption and ensuring reliability in data retrieval and processing. This dual mechanism is crucial for maintaining the robust performance and reliability of the Hadoop ecosystem.

98

Resposta de referência

Situation: We discovered that customer transaction amounts in our data warehouse were incorrectly calculated for three days, affecting executive dashboards. Task: I needed to fix the data, identify the root cause, and prevent future issues while communicating with affected stakeholders. Action: First, I isolated the problem and stopped the pipeline to prevent further corruption. I traced the issue to a currency conversion error in our ETL code. I worked with the data team to reprocess the affected data and created a communication plan to inform stakeholders about the temporary data discrepancy. Result: We restored accurate data within 4 hours and implemented additional validation rules to catch similar issues. I also created a data incident response playbook that reduced our response time for future issues.

99

Resposta de referência

COSHH is an abbreviation that stands for Classification and Optimisation-based Schedule for Heterogeneous Hadoop systems.

100

Resposta de referência

For late-arriving facts: watermarking and reprocessing windows. For late-arriving dimensions: deferred dimension lookups or late-binding joins, and an explicit policy about records that arrive outside the watermark.

101

Resposta de referência

To effectively convey insights to a non-technical audience, simplify the concepts by breaking them down into key components and using relatable analogies. Visual aids like charts or diagrams can enhance understanding, and encouraging questions ensures clarity and engagement.

102

Resposta de referência

For a web app such as a calendar similar to Outlook or Google calendar, collecting data on which calendar views are being used would be worthwhile. The data engineer candidate should analyze and understand the domain they're working in, as collecting data from a calendar web app can differ vastly from collecting data from IoT devices.

103

Resposta de referência

Map dependencies across all pipelines and systems. Validate business-critical pipelines first in the new environment. Run systems in parallel during the transition to ensure consistency. Test for data accuracy and performance. Document the rollout plan and communicate clearly with stakeholders.

104

Resposta de referência

In Excel, the ribbon exists in the topmost area of the window. They contain the toolbars and menu items available in Excel. Ribbons contain multiple tabs, each with its own command set. You can switch the ribbon between shown and hidden using CTRL+F1.

105

Resposta de referência

Providing data quality involves executing data validation inspections, schema validation, de-duplication strategies, and data profiling. Thus, we can log abnormalities and inconsistencies and fixed using predefined rules or manual intervention.

106

Resposta de referência

Event-driven processing is a paradigm where workflows or actions are triggered automatically in response to specific events, such as data updates, file uploads, or system notifications. Example Use Case: Using AWS Lambda to process a CSV file when it is uploaded to an S3 bucket. Lambda triggers an ETL job to parse the file, transform the data, and store it in a database. Benefits: Automation: - Removes manual intervention by triggering workflows based on real-time events. - Example: A database update triggers a notification system to alert users. Scalability: - Handles varying loads by processing events as they occur. - Example: Scaling up functions when there are multiple file uploads. Efficiency: - Resources are used only when events occur, reducing costs. - Example: Serverless architectures like Lambda operate on-demand.

107

Resposta de referência

First check alerts and logs to identify the failure point. Assess business impact and communicate with stakeholders. Apply a short-term fix to restore service if possible, then investigate root cause thoroughly. Implement preventive measures like better monitoring or retry logic.

NÃO QUER PERDER NADA?

Os testes práticos Cisco, PMP, CISA, CISM e AWS 100% aprovados estão à venda!
Obtenha agora

Obtenha uma certificação para destacar o seu currículo.

NÃO QUER PERDER NADA?

Os testes práticos Cisco, PMP, CISA, CISM e AWS 100% aprovados estão à venda! Obtenha agora

Obtenha uma certificação para destacar o seu currículo.

Os testes práticos Cisco, PMP, CISA, CISM e AWS 100% aprovados estão à venda!
Obtenha agora