Spark Developer Associate Exam Sections and Weights: What Each of the Seven Covers

The seven sections at a glance

The weights are published on the Databricks certification page; the exam guide (30 October 2025 edition) lists the 32 objectives under each section. The question estimates are simple arithmetic on the 45 scored questions, not an official per-section count.

Weights from the Databricks certification page; objectives from the official exam guide, verified 5 October 2026. The last column is the weight times 45, rounded; Databricks does not publish per-section question counts.

Thirty-two objectives on 45 questions means most objectives are asked about once and the heavier sections' objectives can come up twice. Certify's full-length mocks sample the sections as 9, 9, 13, 5, 4, 3 and 2, which adds up to exactly 45.

Section 1: Apache Spark Architecture and Components (20%)

Seven objectives, the most of any section, which makes each one worth roughly a question:

• The advantages and challenges of implementing Spark.
• The core components: cluster, driver, worker nodes and executors, CPU cores, memory.
• The architecture: DataFrame and Dataset concepts, the SparkSession lifecycle, caching, storage levels, garbage collection.
• The execution hierarchy: application, job, stage, task.
• Partitioning, including shuffles and partitions.
• Execution patterns: actions, transformations, lazy evaluation.
• The Spark modules: Core, Spark SQL, DataFrames, Pandas API on Spark, Structured Streaming, MLlib.

Know the defaults cold: spark.sql.shuffle.partitions is 200 and cache() on a DataFrame uses MEMORY_AND_DISK. The course covers this in Unit 1 (Chapters 01 to 09), starting with why Spark.

Section 2: Using Spark SQL (20%)

Only four objectives carry the same 20% as Section 1, so each is worth about two questions:

• Reading from and writing to common sources (JDBC, files) with Spark SQL, including overwriting and partitioning by column.
• Running SQL directly on files (ORC, JSON, CSV, text, Delta) and the save modes for output.
• Saving to persistent tables with sorting and partitioning to speed later reads.
• Registering DataFrames as temporary views to query with SQL.

The save modes are a favorite: "append", "overwrite", "ignore" and "errorifexists", the default. Unit 2 (Chapters 10 to 15) covers it, including save modes as a trade-off lesson.

Section 3: Developing DataFrame/DataSet API Applications (30%)

The largest section and the most code-heavy. Ten objectives:

• Manipulating columns, rows and structure: adding, dropping, splitting, renaming, filtering, exploding arrays.
• Deduplication and validation.
• Aggregations such as count, approximate count distinct, mean and summary.
• Dates: Unix epoch to date string, extracting date components.
• Combining DataFrames: inner, left, broadcast and cross joins, multiple keys, union and union all.
• Writing, overwriting and reading DataFrames with schemas.
• Sorting, iterating, printing schemas, converting to lists.
• User-defined functions, with or without stateful operators, including state stores.
• Broadcast variables and accumulators.
• The purpose and implementation of broadcast joins.

Expect four PySpark snippets per question where three use a real method wrongly. Two points that come up: union matches by position and keeps duplicates, and spark.sql.autoBroadcastJoinThreshold defaults to 10 MB. See broadcast hash join versus sort-merge join. Unit 3 (Chapters 16 to 28) covers the section.

Section 4: Troubleshooting and Tuning (10%)

• Performance tuning and cluster utilization: partitioning, repartitioning, coalescing, identifying data skew, reducing shuffles.
• Adaptive Query Execution (AQE) and its benefits.
• Logging and monitoring: driver and executor logs to diagnose out-of-memory errors and underutilization.

AQE has been on by default since Spark 3.2, so it is usually part of the answer. The classic pair here is repartition versus coalesce. Unit 4 is Chapters 29 to 32.

Section 5: Structured Streaming (10%)

• The engine: programming model, micro-batch processing, exactly-once semantics, fault tolerance.
• Creating and writing streaming DataFrames, including output modes and sinks.
• Basic operations: selection, projection, windows, aggregation.
• Streaming deduplication, with and without a watermark.

Output modes are the most testable single decision; see append, update and complete output modes. Unit 5 is Chapters 33 to 36.

Sections 6 and 7: Spark Connect and Pandas API on Spark (5% each)

Two objectives each, about two questions each. Section 6 asks for the features of Spark Connect and the deployment modes (client, cluster, local); Section 7 asks for the advantages of the Pandas API on Spark and how to create and invoke a Pandas UDF. With four objectives behind about four questions, these are the best marks per hour of study on the exam. Local, client and cluster deploy modes covers the deploy-mode objective; the course covers both sections in Units 6 and 7 (Chapters 37 to 40).

How to use the weights

Spend study time roughly in proportion: Sections 1 to 3 are 70% of the exam. But do not skip the small sections. Section 2 has only four objectives for 20% of the marks, and Sections 6 and 7 are narrow enough to master in a few sittings. For the format, cost and a preparation plan, start with the complete Spark Developer Associate guide.

FAQ