# Spring Batch 5 Interview: Partitioning, Chunks and Fault Tolerance

> Ace your Spring Batch 5 interviews: 15 essential questions on partitioning, chunk-oriented processing, and fault tolerance with Java 21 code examples.

- Published: 2026-03-22
- Updated: 2026-03-31
- Author: SharpSkill
- Tags: spring batch, spring boot, java, batch processing, interview questions
- Reading time: 16 min

---

Spring Batch 5 stands as a cornerstone for enterprise-grade data processing in the Spring ecosystem. Technical interviews assess the ability to design robust, scalable, and fault-tolerant batch jobs. Mastering partitioning, chunk-oriented processing, and fault tolerance mechanisms sets senior developers apart.

> **Interview Focus**
>
> Recruiters test deep understanding: why choose partitioning over remote chunking? How to size chunks properly? These architectural decisions reveal real production experience.

## Spring Batch 5 Core Architecture

### Question 1: What are the main components of Spring Batch?

Spring Batch architecture consists of three layers: the Application (jobs and business code), Batch Core (runtime classes to launch and control jobs), and Infrastructure (common readers, writers, and services like RetryTemplate).

```java
// BatchJobConfig.java
// Spring Batch 5 job configuration with Java 21
@Configuration
public class BatchJobConfig {

    // JobRepository stores execution metadata
    // Enables restart and job monitoring
    private final JobRepository jobRepository;
    private final PlatformTransactionManager transactionManager;

    public BatchJobConfig(JobRepository jobRepository,
                          PlatformTransactionManager transactionManager) {
        this.jobRepository = jobRepository;
        this.transactionManager = transactionManager;
    }

    // A Job encapsulates the complete batch process
    // Composed of one or more Steps executed sequentially
    @Bean
    public Job importUserJob(Step processUsersStep, Step cleanupStep) {
        return new JobBuilder("importUserJob", jobRepository)
                .start(processUsersStep)      // Main processing step
                .next(cleanupStep)             // Cleanup step
                .build();
    }

    // A Step represents an independent unit of work
    // Two models: Tasklet (single task) or Chunk (iterative processing)
    @Bean
    public Step processUsersStep(ItemReader<UserRecord> reader,
                                  ItemProcessor<UserRecord, User> processor,
                                  ItemWriter<User> writer) {
        return new StepBuilder("processUsersStep", jobRepository)
                .<UserRecord, User>chunk(100, transactionManager)  // Commit every 100 items
                .reader(reader)       // Reads source data
                .processor(processor) // Transforms each item
                .writer(writer)       // Writes in batches of 100
                .build();
    }
}
```

The JobRepository persists execution state to the database. This persistence enables restarting a failed job exactly where it stopped, without reprocessing already committed data.

### Question 2: What is the difference between Tasklet and Chunk-oriented processing?

**Tasklet** executes a discrete, non-repetitive action: file deletion, stored procedure call, notification email. **Chunk** processes massive volumes by splitting data into manageable batches.

```java
// CleanupTasklet.java
// Tasklet: single action without iteration
@Component
public class CleanupTasklet implements Tasklet {

    private final Path tempDirectory = Path.of("/tmp/batch-work");

    @Override
    public RepeatStatus execute(StepContribution contribution,
                                 ChunkContext chunkContext) throws Exception {
        // Deletes all temporary files from processing
        try (var files = Files.walk(tempDirectory)) {
            files.filter(Files::isRegularFile)
                 .forEach(this::deleteQuietly);
        }

        // FINISHED indicates the tasklet completed its work
        // CONTINUABLE would restart execution (useful for polling)
        return RepeatStatus.FINISHED;
    }

    private void deleteQuietly(Path file) {
        try {
            Files.delete(file);
        } catch (IOException e) {
            // Log and continue - don't fail the job for one file
        }
    }
}
```

```java
// ChunkProcessingConfig.java
// Chunk-oriented: high-volume processing
@Configuration
public class ChunkProcessingConfig {

    @Bean
    public Step processOrdersStep(JobRepository jobRepository,
                                   PlatformTransactionManager transactionManager,
                                   ItemReader<OrderRecord> reader,
                                   ItemProcessor<OrderRecord, ProcessedOrder> processor,
                                   ItemWriter<ProcessedOrder> writer) {
        return new StepBuilder("processOrdersStep", jobRepository)
                // Chunk of 500: reads 500 items, processes, writes, then commits
                .<OrderRecord, ProcessedOrder>chunk(500, transactionManager)
                .reader(reader)
                .processor(processor)
                .writer(writer)
                // Listener to monitor progress
                .listener(new ChunkProgressListener())
                .build();
    }
}
```

Chunk-oriented processing provides critical benefits: optimized memory management (only the current chunk in memory), granular transactions (commit per chunk), and failure recovery at the last committed chunk.

## Deep Dive into Chunk-Oriented Processing

### Question 3: How does the chunk lifecycle work?

Each chunk follows a precise cycle: reading items one by one until reaching the configured size, processing each item individually, then writing the group. A transaction wraps the entire chunk.

```java
// OrderItemReader.java
// ItemReader: reads one item at a time
@StepScope
@Component
public class OrderItemReader implements ItemReader<OrderRecord> {

    // @StepScope: new instance per step execution
    // Enables injecting dynamic job parameters
    @Value("#{jobParameters['startDate']}")
    private LocalDate startDate;

    private Iterator<OrderRecord> orderIterator;

    @BeforeStep
    public void initializeReader(StepExecution stepExecution) {
        // Loads data at step startup
        List<OrderRecord> orders = fetchOrdersFromDate(startDate);
        this.orderIterator = orders.iterator();
    }

    @Override
    public OrderRecord read() {
        // Returns null to signal end of data
        // Spring Batch calls read() until receiving null
        if (orderIterator.hasNext()) {
            return orderIterator.next();
        }
        return null;  // End of dataset
    }

    private List<OrderRecord> fetchOrdersFromDate(LocalDate date) {
        // Fetches from data source
        return List.of();  // Actual implementation
    }
}
```

```java
// OrderItemProcessor.java
// ItemProcessor: transforms each item individually
@Component
public class OrderItemProcessor implements ItemProcessor<OrderRecord, ProcessedOrder> {

    private final PricingService pricingService;
    private final ValidationService validationService;

    public OrderItemProcessor(PricingService pricingService,
                               ValidationService validationService) {
        this.pricingService = pricingService;
        this.validationService = validationService;
    }

    @Override
    public ProcessedOrder process(OrderRecord item) {
        // Returning null filters the item (won't be written)
        if (!validationService.isValid(item)) {
            return null;  // Item filtered
        }

        // Business transformation
        BigDecimal finalPrice = pricingService.calculatePrice(item);

        return new ProcessedOrder(
                item.orderId(),
                item.customerId(),
                finalPrice,
                LocalDateTime.now()
        );
    }
}
```

```java
// OrderItemWriter.java
// ItemWriter: writes the complete chunk in one operation
@Component
public class OrderItemWriter implements ItemWriter<ProcessedOrder> {

    private final JdbcTemplate jdbcTemplate;

    public OrderItemWriter(JdbcTemplate jdbcTemplate) {
        this.jdbcTemplate = jdbcTemplate;
    }

    @Override
    public void write(Chunk<? extends ProcessedOrder> chunk) {
        // The chunk contains all processed items
        // Batch writing for optimized performance
        List<? extends ProcessedOrder> items = chunk.getItems();

        jdbcTemplate.batchUpdate(
                "INSERT INTO processed_orders (order_id, customer_id, final_price, processed_at) VALUES (?, ?, ?, ?)",
                items,
                items.size(),
                (ps, order) -> {
                    ps.setLong(1, order.orderId());
                    ps.setLong(2, order.customerId());
                    ps.setBigDecimal(3, order.finalPrice());
                    ps.setTimestamp(4, Timestamp.valueOf(order.processedAt()));
                }
        );
    }
}
```

If an exception occurs during chunk processing, the transaction rolls back. The job can then resume from that chunk using metadata stored in the JobRepository.

### Question 4: How to choose the optimal chunk size?

Chunk size directly impacts performance and memory consumption. A chunk too small multiplies commits (overhead). A chunk too large consumes excessive memory and lengthens rollbacks on failure.

```java
// ChunkSizingConfig.java
// Dynamic chunk size configuration
@Configuration
public class ChunkSizingConfig {

    // Reasonable default for most cases
    private static final int DEFAULT_CHUNK_SIZE = 100;

    // For lightweight items (few fields)
    private static final int LIGHT_ITEMS_CHUNK_SIZE = 500;

    // For heavyweight items (blobs, documents)
    private static final int HEAVY_ITEMS_CHUNK_SIZE = 25;

    @Bean
    public Step processLightDataStep(JobRepository jobRepository,
                                      PlatformTransactionManager txManager,
                                      ItemReader<LightRecord> reader,
                                      ItemWriter<LightRecord> writer) {
        return new StepBuilder("processLightDataStep", jobRepository)
                // Lightweight items: larger chunks for fewer commits
                .<LightRecord, LightRecord>chunk(LIGHT_ITEMS_CHUNK_SIZE, txManager)
                .reader(reader)
                .writer(writer)
                .build();
    }

    @Bean
    public Step processDocumentsStep(JobRepository jobRepository,
                                      PlatformTransactionManager txManager,
                                      ItemReader<Document> reader,
                                      ItemProcessor<Document, ProcessedDocument> processor,
                                      ItemWriter<ProcessedDocument> writer) {
        return new StepBuilder("processDocumentsStep", jobRepository)
                // Heavy documents: smaller chunks to limit memory
                .<Document, ProcessedDocument>chunk(HEAVY_ITEMS_CHUNK_SIZE, txManager)
                .reader(reader)
                .processor(processor)
                .writer(writer)
                .build();
    }
}
```

> **Rule of Thumb**
>
> Start with 100 items per chunk, then adjust based on metrics: commit time, memory usage, and rollback duration. Use listeners to monitor and identify the sweet spot.

## Partitioning for Parallel Processing

### Question 5: What is partitioning and when should it be used?

Partitioning divides a dataset into independent partitions processed in parallel. Each partition executes in its own thread (local) or on a remote worker. This approach multiplies throughput without sacrificing restartability.

```java
// PartitionedJobConfig.java
// Partitioned job configuration
@Configuration
public class PartitionedJobConfig {

    private final JobRepository jobRepository;
    private final PlatformTransactionManager transactionManager;

    public PartitionedJobConfig(JobRepository jobRepository,
                                 PlatformTransactionManager transactionManager) {
        this.jobRepository = jobRepository;
        this.transactionManager = transactionManager;
    }

    @Bean
    public Job partitionedImportJob(Step partitionedStep) {
        return new JobBuilder("partitionedImportJob", jobRepository)
                .start(partitionedStep)
                .build();
    }

    // Manager step: orchestrates partitions
    @Bean
    public Step partitionedStep(Partitioner partitioner,
                                 Step workerStep,
                                 TaskExecutor taskExecutor) {
        return new StepBuilder("partitionedStep", jobRepository)
                // Divides work via the Partitioner
                .partitioner("workerStep", partitioner)
                // Step executed for each partition
                .step(workerStep)
                // 8 parallel threads
                .taskExecutor(taskExecutor)
                // Number of partitions to create
                .gridSize(8)
                .build();
    }

    // TaskExecutor for parallel execution
    @Bean
    public TaskExecutor batchTaskExecutor() {
        ThreadPoolTaskExecutor executor = new ThreadPoolTaskExecutor();
        executor.setCorePoolSize(8);
        executor.setMaxPoolSize(16);
        executor.setQueueCapacity(50);
        executor.setThreadNamePrefix("batch-partition-");
        executor.initialize();
        return executor;
    }
}
```

```java
// RangePartitioner.java
// Partitioner based on ID ranges
@Component
public class RangePartitioner implements Partitioner {

    private final JdbcTemplate jdbcTemplate;

    public RangePartitioner(JdbcTemplate jdbcTemplate) {
        this.jdbcTemplate = jdbcTemplate;
    }

    @Override
    public Map<String, ExecutionContext> partition(int gridSize) {
        // Retrieves dataset boundaries
        Long minId = jdbcTemplate.queryForObject(
                "SELECT MIN(id) FROM orders WHERE status = 'PENDING'", Long.class);
        Long maxId = jdbcTemplate.queryForObject(
                "SELECT MAX(id) FROM orders WHERE status = 'PENDING'", Long.class);

        if (minId == null || maxId == null) {
            return Map.of();  // No data to process
        }

        // Calculates each partition size
        long range = (maxId - minId) / gridSize + 1;
        Map<String, ExecutionContext> partitions = new HashMap<>();

        for (int i = 0; i < gridSize; i++) {
            ExecutionContext context = new ExecutionContext();
            long start = minId + (i * range);
            long end = Math.min(start + range - 1, maxId);

            // Each partition receives its boundaries
            context.putLong("minId", start);
            context.putLong("maxId", end);
            context.putInt("partitionNumber", i);

            partitions.put("partition" + i, context);
        }

        return partitions;
    }
}
```

Partitioning suits large datasets where items are independent. Partitions must be balanced to prevent a slow partition from slowing down the entire job.

### Question 6: What is the difference between local and remote partitioning?

**Local partitioning** executes all partitions on the same JVM with a thread pool. **Remote partitioning** distributes partitions across multiple JVMs (workers) via messaging middleware.

```java
// RemotePartitioningConfig.java
// Remote partitioning configuration with messaging
@Configuration
public class RemotePartitioningConfig {

    @Bean
    public Step managerStep(JobRepository jobRepository,
                             Partitioner partitioner,
                             MessageChannelPartitionHandler partitionHandler) {
        return new StepBuilder("managerStep", jobRepository)
                .partitioner("workerStep", partitioner)
                // Handler that communicates with remote workers
                .partitionHandler(partitionHandler)
                .build();
    }

    // PartitionHandler sends ExecutionContexts to workers
    @Bean
    public MessageChannelPartitionHandler partitionHandler(
            MessagingTemplate messagingTemplate,
            JobExplorer jobExplorer) {
        MessageChannelPartitionHandler handler = new MessageChannelPartitionHandler();
        handler.setStepName("workerStep");
        handler.setGridSize(4);
        handler.setMessagingOperations(messagingTemplate);
        handler.setJobExplorer(jobExplorer);
        // Timeout waiting for workers to complete
        handler.setPollInterval(5000L);
        return handler;
    }
}
```

```java
// WorkerConfiguration.java
// Worker-side configuration
@Configuration
public class WorkerConfiguration {

    private final JobRepository jobRepository;
    private final PlatformTransactionManager transactionManager;

    public WorkerConfiguration(JobRepository jobRepository,
                                PlatformTransactionManager transactionManager) {
        this.jobRepository = jobRepository;
        this.transactionManager = transactionManager;
    }

    // Worker receives partitions and executes the step
    @Bean
    public Step workerStep(ItemReader<OrderRecord> reader,
                            ItemProcessor<OrderRecord, ProcessedOrder> processor,
                            ItemWriter<ProcessedOrder> writer) {
        return new StepBuilder("workerStep", jobRepository)
                .<OrderRecord, ProcessedOrder>chunk(100, transactionManager)
                // Reader configured with @StepScope to receive partition parameters
                .reader(reader)
                .processor(processor)
                .writer(writer)
                .build();
    }

    // Reader that uses partition boundaries
    @Bean
    @StepScope
    public JdbcCursorItemReader<OrderRecord> partitionedReader(
            DataSource dataSource,
            @Value("#{stepExecutionContext['minId']}") Long minId,
            @Value("#{stepExecutionContext['maxId']}") Long maxId) {
        return new JdbcCursorItemReaderBuilder<OrderRecord>()
                .name("partitionedOrderReader")
                .dataSource(dataSource)
                .sql("SELECT * FROM orders WHERE id BETWEEN ? AND ? AND status = 'PENDING'")
                .preparedStatementSetter(ps -> {
                    ps.setLong(1, minId);
                    ps.setLong(2, maxId);
                })
                .rowMapper(new OrderRecordRowMapper())
                .build();
    }
}
```

## Fault Tolerance and Error Recovery

### Question 7: What fault tolerance mechanisms does Spring Batch offer?

Spring Batch provides three complementary mechanisms: **skip** (ignore failing items), **retry** (automatically retry), and **restart** (resume a failed job). These mechanisms are configured at the step level.

```java
// FaultTolerantStepConfig.java
// Complete fault tolerance configuration
@Configuration
public class FaultTolerantStepConfig {

    @Bean
    public Step faultTolerantStep(JobRepository jobRepository,
                                   PlatformTransactionManager transactionManager,
                                   ItemReader<DataRecord> reader,
                                   ItemProcessor<DataRecord, ProcessedRecord> processor,
                                   ItemWriter<ProcessedRecord> writer,
                                   SkipPolicy customSkipPolicy) {
        return new StepBuilder("faultTolerantStep", jobRepository)
                .<DataRecord, ProcessedRecord>chunk(100, transactionManager)
                .reader(reader)
                .processor(processor)
                .writer(writer)
                // Enables fault tolerant mode
                .faultTolerant()
                // SKIP: ignores up to 10 validation errors
                .skipLimit(10)
                .skip(ValidationException.class)
                .skip(DataIntegrityViolationException.class)
                // Some errors should never be skipped
                .noSkip(FatalBatchException.class)
                // RETRY: retries transient errors
                .retryLimit(3)
                .retry(TransientDataAccessException.class)
                .retry(DeadlockLoserDataAccessException.class)
                // Exponential backoff between retries
                .backOffPolicy(exponentialBackOffPolicy())
                // Listener to log skips
                .listener(skipListener())
                .build();
    }

    @Bean
    public BackOffPolicy exponentialBackOffPolicy() {
        ExponentialBackOffPolicy policy = new ExponentialBackOffPolicy();
        policy.setInitialInterval(1000);  // 1 second
        policy.setMultiplier(2.0);         // Doubles each retry
        policy.setMaxInterval(10000);      // Max 10 seconds
        return policy;
    }

    @Bean
    public SkipListener<DataRecord, ProcessedRecord> skipListener() {
        return new SkipListener<>() {
            @Override
            public void onSkipInRead(Throwable t) {
                // Log unreadable item
            }

            @Override
            public void onSkipInProcess(DataRecord item, Throwable t) {
                // Log item that failed processing
            }

            @Override
            public void onSkipInWrite(ProcessedRecord item, Throwable t) {
                // Log item that failed writing
            }
        };
    }
}
```

Retry suits transient errors (network timeout, database deadlock). Skip suits individual data errors that should not block overall processing.

### Question 8: How to implement a custom SkipPolicy?

A custom SkipPolicy enables fine-grained decision logic: skip based on exception type, error count, or specific business criteria.

```java
// AdaptiveSkipPolicy.java
// SkipPolicy with advanced business logic
@Component
public class AdaptiveSkipPolicy implements SkipPolicy {

    private static final int MAX_SKIP_COUNT = 100;
    private static final double MAX_SKIP_PERCENTAGE = 0.05;  // 5% max

    private final AtomicInteger totalProcessed = new AtomicInteger(0);
    private final AtomicInteger skipCount = new AtomicInteger(0);

    @Override
    public boolean shouldSkip(Throwable exception, long skipCountSoFar) {
        // Never skip fatal errors
        if (exception instanceof FatalBatchException
                || exception instanceof OutOfMemoryError) {
            return false;
        }

        // Absolute skip limit
        if (skipCountSoFar >= MAX_SKIP_COUNT) {
            return false;  // Stop the job
        }

        // Percentage limit
        int total = totalProcessed.get();
        if (total > 1000) {  // Apply only after warmup
            double skipPercentage = (double) skipCountSoFar / total;
            if (skipPercentage > MAX_SKIP_PERCENTAGE) {
                return false;  // Too many errors proportionally
            }
        }

        // Skip validation and data errors
        return exception instanceof ValidationException
                || exception instanceof DataFormatException
                || exception instanceof IllegalArgumentException;
    }

    // Called by a listener to track progress
    public void incrementProcessed() {
        totalProcessed.incrementAndGet();
    }
}
```

### Question 9: How does restarting a failed job work?

The JobRepository stores each execution's state. On restart, Spring Batch identifies the last committed chunk and resumes from that point. Successfully processed items are not reprocessed.

```java
// JobRestartService.java
// Job restart management service
@Service
public class JobRestartService {

    private final JobLauncher jobLauncher;
    private final JobExplorer jobExplorer;
    private final JobRepository jobRepository;
    private final Job importJob;

    public JobRestartService(JobLauncher jobLauncher,
                              JobExplorer jobExplorer,
                              JobRepository jobRepository,
                              @Qualifier("importJob") Job importJob) {
        this.jobLauncher = jobLauncher;
        this.jobExplorer = jobExplorer;
        this.jobRepository = jobRepository;
        this.importJob = importJob;
    }

    public JobExecution restartFailedJob(Long jobExecutionId) throws Exception {
        // Retrieves the failed execution
        JobExecution failedExecution = jobExplorer.getJobExecution(jobExecutionId);

        if (failedExecution == null) {
            throw new IllegalArgumentException("Job execution not found: " + jobExecutionId);
        }

        // Verifies the job can be restarted
        if (!failedExecution.getStatus().equals(BatchStatus.FAILED)) {
            throw new IllegalStateException("Only FAILED jobs can be restarted");
        }

        // Uses the same parameters as the original execution
        JobParameters originalParams = failedExecution.getJobParameters();

        // Relaunches the job - automatically resumes from last checkpoint
        return jobLauncher.run(importJob, originalParams);
    }

    public List<JobExecution> findRestartableJobs() {
        // Lists all FAILED executions not yet restarted
        return jobExplorer.findJobInstancesByJobName(importJob.getName(), 0, 100)
                .stream()
                .flatMap(instance -> jobExplorer.getJobExecutions(instance).stream())
                .filter(exec -> exec.getStatus() == BatchStatus.FAILED)
                .filter(this::isRestartable)
                .toList();
    }

    private boolean isRestartable(JobExecution execution) {
        // Verifies no more recent successful execution exists
        JobInstance instance = execution.getJobInstance();
        return jobExplorer.getJobExecutions(instance).stream()
                .noneMatch(exec -> exec.getStatus() == BatchStatus.COMPLETED);
    }
}
```

> **Interview Pitfall**
>
> A job can only be restarted if JobParameters are identical. Modifying a parameter creates a new job instance, losing the progress history.

## Scaling and Optimization

### Question 10: What scaling strategies are available?

Spring Batch offers four strategies: **multi-threaded step** (multiple threads read in parallel), **parallel steps** (independent steps in parallel), **remote chunking** (distributed processing), and **partitioning** (distributed data).

```java
// MultiThreadedStepConfig.java
// Multi-threaded step: multiple threads process the same dataset
@Configuration
public class MultiThreadedStepConfig {

    @Bean
    public Step multiThreadedStep(JobRepository jobRepository,
                                   PlatformTransactionManager transactionManager,
                                   ItemReader<Record> reader,
                                   ItemProcessor<Record, ProcessedRecord> processor,
                                   ItemWriter<ProcessedRecord> writer,
                                   TaskExecutor taskExecutor) {
        return new StepBuilder("multiThreadedStep", jobRepository)
                .<Record, ProcessedRecord>chunk(100, transactionManager)
                // CAUTION: reader must be thread-safe
                .reader(synchronizedReader(reader))
                .processor(processor)
                .writer(writer)
                // 4 threads process chunks in parallel
                .taskExecutor(taskExecutor)
                .throttleLimit(4)
                .build();
    }

    // Wrapper to make the reader thread-safe
    private ItemReader<Record> synchronizedReader(ItemReader<Record> reader) {
        SynchronizedItemStreamReader<Record> syncReader = new SynchronizedItemStreamReader<>();
        syncReader.setDelegate((ItemStreamReader<Record>) reader);
        return syncReader;
    }
}
```

```java
// ParallelStepsConfig.java
// Executing independent steps in parallel
@Configuration
public class ParallelStepsConfig {

    @Bean
    public Job parallelJob(JobRepository jobRepository,
                            Step loadCustomersStep,
                            Step loadProductsStep,
                            Step loadOrdersStep,
                            Step processDataStep) {
        // Parallel flow: customers and products loaded simultaneously
        Flow loadCustomersFlow = new FlowBuilder<Flow>("loadCustomersFlow")
                .start(loadCustomersStep)
                .build();

        Flow loadProductsFlow = new FlowBuilder<Flow>("loadProductsFlow")
                .start(loadProductsStep)
                .build();

        Flow loadOrdersFlow = new FlowBuilder<Flow>("loadOrdersFlow")
                .start(loadOrdersStep)
                .build();

        // Split executes flows in parallel
        return new JobBuilder("parallelJob", jobRepository)
                .start(new FlowBuilder<Flow>("parallelLoadFlow")
                        .split(new SimpleAsyncTaskExecutor())
                        .add(loadCustomersFlow, loadProductsFlow, loadOrdersFlow)
                        .build())
                // After parallel loading, sequential processing
                .next(processDataStep)
                .build()
                .build();
    }
}
```

Multi-threading suits cases where the reader can be synchronized. Partitioning is preferred for large volumes since each partition has its own reader without contention.

### Question 11: How to monitor job performance?

Spring Batch exposes metrics via listeners and JobRepository. Integration with Micrometer enables export to Prometheus, Grafana, or other monitoring systems.

```java
// BatchMetricsConfig.java
// Monitoring configuration with Micrometer
@Configuration
public class BatchMetricsConfig {

    private final MeterRegistry meterRegistry;

    public BatchMetricsConfig(MeterRegistry meterRegistry) {
        this.meterRegistry = meterRegistry;
    }

    @Bean
    public JobExecutionListener metricsJobListener() {
        return new JobExecutionListener() {

            private Timer.Sample jobTimer;

            @Override
            public void beforeJob(JobExecution jobExecution) {
                // Starts the job duration timer
                jobTimer = Timer.start(meterRegistry);
                Counter.builder("batch.job.started")
                        .tag("job", jobExecution.getJobInstance().getJobName())
                        .register(meterRegistry)
                        .increment();
            }

            @Override
            public void afterJob(JobExecution jobExecution) {
                // Records total duration
                jobTimer.stop(Timer.builder("batch.job.duration")
                        .tag("job", jobExecution.getJobInstance().getJobName())
                        .tag("status", jobExecution.getStatus().toString())
                        .register(meterRegistry));

                // Job counter by status
                Counter.builder("batch.job.completed")
                        .tag("job", jobExecution.getJobInstance().getJobName())
                        .tag("status", jobExecution.getStatus().toString())
                        .register(meterRegistry)
                        .increment();
            }
        };
    }

    @Bean
    public StepExecutionListener metricsStepListener() {
        return new StepExecutionListener() {

            @Override
            public void afterStep(StepExecution stepExecution) {
                String jobName = stepExecution.getJobExecution().getJobInstance().getJobName();
                String stepName = stepExecution.getStepName();

                // Throughput metrics
                Gauge.builder("batch.step.read.count", stepExecution, StepExecution::getReadCount)
                        .tag("job", jobName)
                        .tag("step", stepName)
                        .register(meterRegistry);

                Gauge.builder("batch.step.write.count", stepExecution, StepExecution::getWriteCount)
                        .tag("job", jobName)
                        .tag("step", stepName)
                        .register(meterRegistry);

                Gauge.builder("batch.step.skip.count", stepExecution, StepExecution::getSkipCount)
                        .tag("job", jobName)
                        .tag("step", stepName)
                        .register(meterRegistry);

                return null;
            }
        };
    }
}
```

### Question 12: What are common pitfalls with partitioning?

Frequent mistakes include: unbalanced partitions (one partition contains 90% of data), non-thread-safe readers, and incorrect state management between partitions.

```java
// BalancedPartitioner.java
// Partitioner that actually balances the load
@Component
public class BalancedPartitioner implements Partitioner {

    private final JdbcTemplate jdbcTemplate;

    public BalancedPartitioner(JdbcTemplate jdbcTemplate) {
        this.jdbcTemplate = jdbcTemplate;
    }

    @Override
    public Map<String, ExecutionContext> partition(int gridSize) {
        // Counts total items to process
        Integer totalCount = jdbcTemplate.queryForObject(
                "SELECT COUNT(*) FROM orders WHERE status = 'PENDING'", Integer.class);

        if (totalCount == null || totalCount == 0) {
            return Map.of();
        }

        // Calculates target size per partition
        int itemsPerPartition = (int) Math.ceil((double) totalCount / gridSize);

        Map<String, ExecutionContext> partitions = new HashMap<>();

        // Uses OFFSET/LIMIT for balanced partitions
        // More expensive than ranges but guarantees balance
        for (int i = 0; i < gridSize; i++) {
            ExecutionContext context = new ExecutionContext();
            context.putInt("offset", i * itemsPerPartition);
            context.putInt("limit", itemsPerPartition);
            context.putInt("partitionNumber", i);

            partitions.put("partition" + i, context);
        }

        return partitions;
    }
}

// OffsetBasedReader.java
// Reader compatible with offset-based partitioning
@StepScope
@Component
public class OffsetBasedReader implements ItemReader<OrderRecord>, ItemStream {

    private final JdbcTemplate jdbcTemplate;
    private Iterator<OrderRecord> iterator;

    @Value("#{stepExecutionContext['offset']}")
    private int offset;

    @Value("#{stepExecutionContext['limit']}")
    private int limit;

    public OffsetBasedReader(JdbcTemplate jdbcTemplate) {
        this.jdbcTemplate = jdbcTemplate;
    }

    @Override
    public void open(ExecutionContext executionContext) {
        // Loads exactly the portion assigned to this partition
        List<OrderRecord> records = jdbcTemplate.query(
                "SELECT * FROM orders WHERE status = 'PENDING' ORDER BY id LIMIT ? OFFSET ?",
                new OrderRecordRowMapper(),
                limit, offset
        );
        this.iterator = records.iterator();
    }

    @Override
    public OrderRecord read() {
        return iterator.hasNext() ? iterator.next() : null;
    }

    @Override
    public void update(ExecutionContext executionContext) {
        // State saving for restart if needed
    }

    @Override
    public void close() {
        // Cleanup
    }
}
```

## Advanced Questions for Seniors

### Question 13: How to handle dependencies between jobs?

Spring Batch doesn't natively manage inter-job dependencies. Solutions include: external orchestrators (Airflow, Kubernetes CronJob), or custom implementation with JobExplorer.

```java
// JobDependencyService.java
// Inter-job dependency management
@Service
public class JobDependencyService {

    private final JobExplorer jobExplorer;
    private final JobLauncher jobLauncher;
    private final Map<String, Job> jobs;

    public JobDependencyService(JobExplorer jobExplorer,
                                  JobLauncher jobLauncher,
                                  Map<String, Job> jobs) {
        this.jobExplorer = jobExplorer;
        this.jobLauncher = jobLauncher;
        this.jobs = jobs;
    }

    public JobExecution runWithDependencies(String jobName,
                                             JobParameters params,
                                             List<String> dependsOn) throws Exception {
        // Verifies all dependencies succeeded
        for (String dependency : dependsOn) {
            if (!hasSuccessfulExecution(dependency, params)) {
                throw new JobExecutionException(
                        "Dependency not satisfied: " + dependency);
            }
        }

        Job job = jobs.get(jobName);
        if (job == null) {
            throw new IllegalArgumentException("Unknown job: " + jobName);
        }

        return jobLauncher.run(job, params);
    }

    private boolean hasSuccessfulExecution(String jobName, JobParameters params) {
        // Looks for a COMPLETED execution with the same business parameters
        return jobExplorer.findJobInstancesByJobName(jobName, 0, 1)
                .stream()
                .flatMap(instance -> jobExplorer.getJobExecutions(instance).stream())
                .filter(exec -> exec.getStatus() == BatchStatus.COMPLETED)
                .anyMatch(exec -> matchesBusinessParams(exec.getJobParameters(), params));
    }

    private boolean matchesBusinessParams(JobParameters actual, JobParameters expected) {
        // Compares business parameters (ignores execution timestamps)
        String actualDate = actual.getString("businessDate");
        String expectedDate = expected.getString("businessDate");
        return Objects.equals(actualDate, expectedDate);
    }
}
```

### Question 14: How to effectively test a Spring Batch job?

Testing Spring Batch jobs requires a layered approach: unit tests for components (reader, processor, writer), integration tests for steps, and end-to-end tests for complete jobs.

```java
// OrderProcessorTest.java
// Processor unit test
@ExtendWith(MockitoExtension.class)
class OrderProcessorTest {

    @Mock
    private PricingService pricingService;

    @Mock
    private ValidationService validationService;

    @InjectMocks
    private OrderItemProcessor processor;

    @Test
    void shouldProcessValidOrder() {
        // Given
        OrderRecord input = new OrderRecord(1L, 100L, BigDecimal.TEN);
        when(validationService.isValid(input)).thenReturn(true);
        when(pricingService.calculatePrice(input)).thenReturn(new BigDecimal("12.50"));

        // When
        ProcessedOrder result = processor.process(input);

        // Then
        assertThat(result).isNotNull();
        assertThat(result.finalPrice()).isEqualTo(new BigDecimal("12.50"));
    }

    @Test
    void shouldFilterInvalidOrder() {
        // Given
        OrderRecord input = new OrderRecord(1L, 100L, BigDecimal.TEN);
        when(validationService.isValid(input)).thenReturn(false);

        // When
        ProcessedOrder result = processor.process(input);

        // Then - null means filtered
        assertThat(result).isNull();
        verify(pricingService, never()).calculatePrice(any());
    }
}
```

```java
// ImportJobIntegrationTest.java
// Complete job integration test
@SpringBatchTest
@SpringBootTest
@ActiveProfiles("test")
class ImportJobIntegrationTest {

    @Autowired
    private JobLauncherTestUtils jobLauncherTestUtils;

    @Autowired
    private JobRepositoryTestUtils jobRepositoryTestUtils;

    @Autowired
    private JdbcTemplate jdbcTemplate;

    @BeforeEach
    void setup() {
        // Cleans metadata between tests
        jobRepositoryTestUtils.removeJobExecutions();
        // Resets test data
        jdbcTemplate.execute("DELETE FROM processed_orders");
        jdbcTemplate.execute("DELETE FROM orders");
    }

    @Test
    void shouldCompleteJobSuccessfully() throws Exception {
        // Given - test data
        insertTestOrders(100);

        // When
        JobParameters params = new JobParametersBuilder()
                .addLocalDate("businessDate", LocalDate.now())
                .addLong("run.id", System.currentTimeMillis())
                .toJobParameters();

        JobExecution execution = jobLauncherTestUtils.launchJob(params);

        // Then
        assertThat(execution.getStatus()).isEqualTo(BatchStatus.COMPLETED);
        assertThat(countProcessedOrders()).isEqualTo(100);
    }

    @Test
    void shouldHandleEmptyDataset() throws Exception {
        // Given - no data

        // When
        JobExecution execution = jobLauncherTestUtils.launchJob();

        // Then - job succeeds even without data
        assertThat(execution.getStatus()).isEqualTo(BatchStatus.COMPLETED);
    }

    @Test
    void shouldRestartFromFailurePoint() throws Exception {
        // Given - simulates mid-processing error
        insertTestOrders(100);
        insertPoisonOrder(50);  // Causes an error

        // When - first execution fails
        JobExecution firstExecution = jobLauncherTestUtils.launchJob();
        assertThat(firstExecution.getStatus()).isEqualTo(BatchStatus.FAILED);

        // Fix the data
        removePoisonOrder(50);

        // When - restart
        JobExecution restartExecution = jobLauncherTestUtils.launchJob(
                firstExecution.getJobParameters());

        // Then - resumes from failure point
        assertThat(restartExecution.getStatus()).isEqualTo(BatchStatus.COMPLETED);
    }

    private void insertTestOrders(int count) {
        for (int i = 1; i <= count; i++) {
            jdbcTemplate.update(
                    "INSERT INTO orders (id, customer_id, amount, status) VALUES (?, ?, ?, 'PENDING')",
                    i, i * 10, BigDecimal.valueOf(i * 10));
        }
    }

    private int countProcessedOrders() {
        return jdbcTemplate.queryForObject(
                "SELECT COUNT(*) FROM processed_orders", Integer.class);
    }
}
```

### Question 15: How to optimize database write performance?

Writing often becomes the bottleneck. Optimizations include: JDBC batch inserts, disabling constraints during loading, and using staging tables.

```java
// OptimizedJdbcWriter.java
// Writer optimized for high volumes
@Component
public class OptimizedJdbcWriter implements ItemWriter<ProcessedOrder> {

    private final JdbcTemplate jdbcTemplate;
    private final DataSource dataSource;

    public OptimizedJdbcWriter(JdbcTemplate jdbcTemplate, DataSource dataSource) {
        this.jdbcTemplate = jdbcTemplate;
        this.dataSource = dataSource;
    }

    @Override
    public void write(Chunk<? extends ProcessedOrder> chunk) throws Exception {
        List<? extends ProcessedOrder> items = chunk.getItems();

        if (items.isEmpty()) {
            return;
        }

        // Uses PreparedStatement with batch
        try (Connection connection = dataSource.getConnection();
             PreparedStatement ps = connection.prepareStatement(
                     "INSERT INTO processed_orders (order_id, customer_id, final_price, processed_at) " +
                             "VALUES (?, ?, ?, ?)")) {

            for (ProcessedOrder order : items) {
                ps.setLong(1, order.orderId());
                ps.setLong(2, order.customerId());
                ps.setBigDecimal(3, order.finalPrice());
                ps.setTimestamp(4, Timestamp.valueOf(order.processedAt()));
                ps.addBatch();
            }

            // Executes all inserts in a single network operation
            ps.executeBatch();
        }
    }
}

// StagingTableWriter.java
// Staging table pattern for very large volumes
@Component
public class StagingTableWriter implements ItemWriter<ProcessedOrder>, StepExecutionListener {

    private final JdbcTemplate jdbcTemplate;
    private String stagingTable;

    public StagingTableWriter(JdbcTemplate jdbcTemplate) {
        this.jdbcTemplate = jdbcTemplate;
    }

    @Override
    public void beforeStep(StepExecution stepExecution) {
        // Creates a temporary table for this step
        stagingTable = "staging_orders_" + stepExecution.getId();
        jdbcTemplate.execute(
                "CREATE TEMP TABLE " + stagingTable + " (LIKE processed_orders INCLUDING ALL)");
    }

    @Override
    public void write(Chunk<? extends ProcessedOrder> chunk) {
        // Writes to staging table (without FK constraints)
        String sql = "INSERT INTO " + stagingTable +
                " (order_id, customer_id, final_price, processed_at) VALUES (?, ?, ?, ?)";

        jdbcTemplate.batchUpdate(sql, chunk.getItems(), chunk.size(),
                (ps, order) -> {
                    ps.setLong(1, order.orderId());
                    ps.setLong(2, order.customerId());
                    ps.setBigDecimal(3, order.finalPrice());
                    ps.setTimestamp(4, Timestamp.valueOf(order.processedAt()));
                });
    }

    @Override
    public ExitStatus afterStep(StepExecution stepExecution) {
        if (stepExecution.getStatus() == BatchStatus.COMPLETED) {
            // Bulk copy to final table
            jdbcTemplate.execute(
                    "INSERT INTO processed_orders SELECT * FROM " + stagingTable);
        }
        // Cleans up staging table
        jdbcTemplate.execute("DROP TABLE IF EXISTS " + stagingTable);
        return stepExecution.getExitStatus();
    }
}
```

## Conclusion

Mastering Spring Batch 5 in technical interviews relies on deep understanding of internal mechanisms:

✅ **Architecture**: Job → Step → Chunk (Reader, Processor, Writer)

✅ **Chunk processing**: sizing, lifecycle, transactions

✅ **Partitioning**: local vs remote, partition balancing

✅ **Fault tolerance**: skip, retry, restart with appropriate policy

✅ **Scaling**: multi-threading, parallel steps, remote chunking

✅ **Testing**: unit, integration, end-to-end

✅ **Optimization**: batch writes, staging tables, monitoring

Advanced questions test the ability to justify architectural choices based on context: data volume, time constraints, error tolerance, and available infrastructure.

---

Source: SharpSkill (https://sharpskill.dev), tech interview preparation for your real stack.
HTML version of this page: https://sharpskill.dev/en/blog/spring-boot/spring-batch-5-interview-partitioning-chunks-fault-tolerance