Entrevista Spring Batch 5: Particionamiento, Chunks y Tolerancia

// Configuración de un job Spring Batch 5 con Java 21
@Configuration
public class BatchJobConfig {

    // JobRepository almacena los metadatos de ejecución
    // Permite el reinicio y monitoreo de jobs
    private final JobRepository jobRepository;
    private final PlatformTransactionManager transactionManager;

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

    // Un Job encapsula el proceso batch completo
    // Compuesto por uno o más Steps ejecutados secuencialmente
    @Bean
    public Job importUserJob(Step processUsersStep, Step cleanupStep) {
        return new JobBuilder("importUserJob", jobRepository)
                .start(processUsersStep)      // Step principal de procesamiento
                .next(cleanupStep)             // Step de limpieza
                .build();
    }

    // Un Step representa una unidad de trabajo independiente
    // Dos modelos: Tasklet (tarea única) o Chunk (procesamiento iterativo)
    @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 cada 100 items
                .reader(reader)       // Lee los datos fuente
                .processor(processor) // Transforma cada item
                .writer(writer)       // Escribe por lotes de 100
                .build();
    }
}

// Tasklet: acción única sin iteración
@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 {
        // Elimina todos los archivos temporales del procesamiento
        try (var files = Files.walk(tempDirectory)) {
            files.filter(Files::isRegularFile)
                 .forEach(this::deleteQuietly);
        }

        // FINISHED indica que el tasklet completó su trabajo
        // CONTINUABLE reiniciaría la ejecución (útil para polling)
        return RepeatStatus.FINISHED;
    }

    private void deleteQuietly(Path file) {
        try {
            Files.delete(file);
        } catch (IOException e) {
            // Loguea y continúa - no falla el job por un archivo
        }
    }
}

// Procesamiento por chunks: procesamiento de alto volumen
@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 de 500: lee 500 items, procesa, escribe y luego commit
                .<OrderRecord, ProcessedOrder>chunk(500, transactionManager)
                .reader(reader)
                .processor(processor)
                .writer(writer)
                // Listener para monitorear el progreso
                .listener(new ChunkProgressListener())
                .build();
    }
}

// ItemReader: lee un item a la vez
@StepScope
@Component
public class OrderItemReader implements ItemReader<OrderRecord> {

    // @StepScope: nueva instancia por ejecución del step
    // Permite inyectar parámetros dinámicos del job
    @Value("#{jobParameters['startDate']}")
    private LocalDate startDate;

    private Iterator<OrderRecord> orderIterator;

    @BeforeStep
    public void initializeReader(StepExecution stepExecution) {
        // Carga los datos al inicio del step
        List<OrderRecord> orders = fetchOrdersFromDate(startDate);
        this.orderIterator = orders.iterator();
    }

    @Override
    public OrderRecord read() {
        // Devuelve null para señalar el fin de los datos
        // Spring Batch llama a read() hasta recibir null
        if (orderIterator.hasNext()) {
            return orderIterator.next();
        }
        return null;  // Fin del dataset
    }

    private List<OrderRecord> fetchOrdersFromDate(LocalDate date) {
        // Recupera desde la fuente de datos
        return List.of();  // Implementación real
    }
}

// ItemProcessor: transforma cada item individualmente
@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) {
        // Devolver null filtra el item (no será escrito)
        if (!validationService.isValid(item)) {
            return null;  // Item filtrado
        }

        // Transformación de negocio
        BigDecimal finalPrice = pricingService.calculatePrice(item);

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

// ItemWriter: escribe el chunk completo en una sola operación
@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) {
        // El chunk contiene todos los items procesados
        // Escritura por lotes para rendimiento optimizado
        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()));
                }
        );
    }
}

// Configuración dinámica del tamaño de chunk
@Configuration
public class ChunkSizingConfig {

    // Valor por defecto razonable para la mayoría de casos
    private static final int DEFAULT_CHUNK_SIZE = 100;

    // Para items ligeros (pocos campos)
    private static final int LIGHT_ITEMS_CHUNK_SIZE = 500;

    // Para items pesados (blobs, documentos)
    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)
                // Items ligeros: chunks más grandes para menos 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)
                // Documentos pesados: chunks más pequeños para limitar memoria
                .<Document, ProcessedDocument>chunk(HEAVY_ITEMS_CHUNK_SIZE, txManager)
                .reader(reader)
                .processor(processor)
                .writer(writer)
                .build();
    }
}

// Configuración de un job particionado
@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();
    }

    // Step manager: orquesta las particiones
    @Bean
    public Step partitionedStep(Partitioner partitioner,
                                 Step workerStep,
                                 TaskExecutor taskExecutor) {
        return new StepBuilder("partitionedStep", jobRepository)
                // Divide el trabajo vía el Partitioner
                .partitioner("workerStep", partitioner)
                // Step ejecutado para cada partición
                .step(workerStep)
                // 8 threads paralelos
                .taskExecutor(taskExecutor)
                // Número de particiones a crear
                .gridSize(8)
                .build();
    }

    // TaskExecutor para ejecución paralela
    @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;
    }
}

// Partitioner basado en rangos de ID
@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) {
        // Recupera los límites del dataset
        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();  // Sin datos para procesar
        }

        // Calcula el tamaño de cada partición
        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);

            // Cada partición recibe sus límites
            context.putLong("minId", start);
            context.putLong("maxId", end);
            context.putInt("partitionNumber", i);

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

        return partitions;
    }
}

// Configuración de particionamiento remoto con mensajería
@Configuration
public class RemotePartitioningConfig {

    @Bean
    public Step managerStep(JobRepository jobRepository,
                             Partitioner partitioner,
                             MessageChannelPartitionHandler partitionHandler) {
        return new StepBuilder("managerStep", jobRepository)
                .partitioner("workerStep", partitioner)
                // Handler que comunica con los workers remotos
                .partitionHandler(partitionHandler)
                .build();
    }

    // PartitionHandler envía los ExecutionContexts a los 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 de espera para que los workers terminen
        handler.setPollInterval(5000L);
        return handler;
    }
}

// Configuración del lado worker
@Configuration
public class WorkerConfiguration {

    private final JobRepository jobRepository;
    private final PlatformTransactionManager transactionManager;

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

    // El worker recibe particiones y ejecuta el 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 configurado con @StepScope para recibir parámetros de partición
                .reader(reader)
                .processor(processor)
                .writer(writer)
                .build();
    }

    // Reader que usa los límites de partición
    @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();
    }
}

// Configuración completa de tolerancia a fallos
@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)
                // Activa el modo tolerante a fallos
                .faultTolerant()
                // SKIP: ignora hasta 10 errores de validación
                .skipLimit(10)
                .skip(ValidationException.class)
                .skip(DataIntegrityViolationException.class)
                // Algunos errores nunca deben ser ignorados
                .noSkip(FatalBatchException.class)
                // RETRY: reintenta errores transitorios
                .retryLimit(3)
                .retry(TransientDataAccessException.class)
                .retry(DeadlockLoserDataAccessException.class)
                // Backoff exponencial entre reintentos
                .backOffPolicy(exponentialBackOffPolicy())
                // Listener para loguear los skips
                .listener(skipListener())
                .build();
    }

    @Bean
    public BackOffPolicy exponentialBackOffPolicy() {
        ExponentialBackOffPolicy policy = new ExponentialBackOffPolicy();
        policy.setInitialInterval(1000);  // 1 segundo
        policy.setMultiplier(2.0);         // Se duplica en cada reintento
        policy.setMaxInterval(10000);      // Máximo 10 segundos
        return policy;
    }

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

            @Override
            public void onSkipInProcess(DataRecord item, Throwable t) {
                // Loguea el item que falló al procesar
            }

            @Override
            public void onSkipInWrite(ProcessedRecord item, Throwable t) {
                // Loguea el item que falló al escribir
            }
        };
    }
}

// SkipPolicy con lógica de negocio avanzada
@Component
public class AdaptiveSkipPolicy implements SkipPolicy {

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

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

    @Override
    public boolean shouldSkip(Throwable exception, long skipCountSoFar) {
        // Nunca ignorar errores fatales
        if (exception instanceof FatalBatchException
                || exception instanceof OutOfMemoryError) {
            return false;
        }

        // Límite absoluto de skips
        if (skipCountSoFar >= MAX_SKIP_COUNT) {
            return false;  // Detiene el job
        }

        // Límite porcentual
        int total = totalProcessed.get();
        if (total > 1000) {  // Aplicar solo después del warmup
            double skipPercentage = (double) skipCountSoFar / total;
            if (skipPercentage > MAX_SKIP_PERCENTAGE) {
                return false;  // Demasiados errores proporcionalmente
            }
        }

        // Ignora errores de validación y datos
        return exception instanceof ValidationException
                || exception instanceof DataFormatException
                || exception instanceof IllegalArgumentException;
    }

    // Llamado por un listener para rastrear el progreso
    public void incrementProcessed() {
        totalProcessed.incrementAndGet();
    }
}

// Servicio de gestión de reinicio de jobs
@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 {
        // Recupera la ejecución fallida
        JobExecution failedExecution = jobExplorer.getJobExecution(jobExecutionId);

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

        // Verifica que el job pueda reiniciarse
        if (!failedExecution.getStatus().equals(BatchStatus.FAILED)) {
            throw new IllegalStateException("Only FAILED jobs can be restarted");
        }

        // Usa los mismos parámetros que la ejecución original
        JobParameters originalParams = failedExecution.getJobParameters();

        // Relanza el job - reanuda automáticamente desde el último checkpoint
        return jobLauncher.run(importJob, originalParams);
    }

    public List<JobExecution> findRestartableJobs() {
        // Lista todas las ejecuciones FAILED no reiniciadas aún
        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) {
        // Verifica que no exista ejecución exitosa más reciente
        JobInstance instance = execution.getJobInstance();
        return jobExplorer.getJobExecutions(instance).stream()
                .noneMatch(exec -> exec.getStatus() == BatchStatus.COMPLETED);
    }
}

// Step multi-threaded: varios threads procesan el mismo 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)
                // CUIDADO: el reader debe ser thread-safe
                .reader(synchronizedReader(reader))
                .processor(processor)
                .writer(writer)
                // 4 threads procesan chunks en paralelo
                .taskExecutor(taskExecutor)
                .throttleLimit(4)
                .build();
    }

    // Wrapper para hacer el reader thread-safe
    private ItemReader<Record> synchronizedReader(ItemReader<Record> reader) {
        SynchronizedItemStreamReader<Record> syncReader = new SynchronizedItemStreamReader<>();
        syncReader.setDelegate((ItemStreamReader<Record>) reader);
        return syncReader;
    }
}

// Ejecución de steps independientes en paralelo
@Configuration
public class ParallelStepsConfig {

    @Bean
    public Job parallelJob(JobRepository jobRepository,
                            Step loadCustomersStep,
                            Step loadProductsStep,
                            Step loadOrdersStep,
                            Step processDataStep) {
        // Flujo paralelo: customers y products cargados simultáneamente
        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 ejecuta los flujos en paralelo
        return new JobBuilder("parallelJob", jobRepository)
                .start(new FlowBuilder<Flow>("parallelLoadFlow")
                        .split(new SimpleAsyncTaskExecutor())
                        .add(loadCustomersFlow, loadProductsFlow, loadOrdersFlow)
                        .build())
                // Después de la carga paralela, procesamiento secuencial
                .next(processDataStep)
                .build()
                .build();
    }
}

// Configuración de monitoreo con 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) {
                // Inicia el timer de duración del job
                jobTimer = Timer.start(meterRegistry);
                Counter.builder("batch.job.started")
                        .tag("job", jobExecution.getJobInstance().getJobName())
                        .register(meterRegistry)
                        .increment();
            }

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

                // Contador de jobs por estado
                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();

                // Métricas de throughput
                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;
            }
        };
    }
}

// Partitioner que realmente balancea la carga
@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) {
        // Cuenta el total de items a procesar
        Integer totalCount = jdbcTemplate.queryForObject(
                "SELECT COUNT(*) FROM orders WHERE status = 'PENDING'", Integer.class);

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

        // Calcula el tamaño objetivo por partición
        int itemsPerPartition = (int) Math.ceil((double) totalCount / gridSize);

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

        // Usa OFFSET/LIMIT para particiones balanceadas
        // Más costoso que rangos pero garantiza el 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 con particionamiento basado en offset
@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) {
        // Carga exactamente la porción asignada a esta partición
        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) {
        // Guardado de estado para reinicio si es necesario
    }

    @Override
    public void close() {
        // Limpieza
    }
}

// Gestión de dependencias entre jobs
@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 {
        // Verifica que todas las dependencias hayan tenido éxito
        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) {
        // Busca una ejecución COMPLETED con los mismos parámetros de negocio
        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) {
        // Compara los parámetros de negocio (ignora timestamps de ejecución)
        String actualDate = actual.getString("businessDate");
        String expectedDate = expected.getString("businessDate");
        return Objects.equals(actualDate, expectedDate);
    }
}

// Test unitario del processor
@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 significa filtrado
        assertThat(result).isNull();
        verify(pricingService, never()).calculatePrice(any());
    }
}

// Test de integración del job completo
@SpringBatchTest
@SpringBootTest
@ActiveProfiles("test")
class ImportJobIntegrationTest {

    @Autowired
    private JobLauncherTestUtils jobLauncherTestUtils;

    @Autowired
    private JobRepositoryTestUtils jobRepositoryTestUtils;

    @Autowired
    private JdbcTemplate jdbcTemplate;

    @BeforeEach
    void setup() {
        // Limpia los metadatos entre tests
        jobRepositoryTestUtils.removeJobExecutions();
        // Resetea los datos de test
        jdbcTemplate.execute("DELETE FROM processed_orders");
        jdbcTemplate.execute("DELETE FROM orders");
    }

    @Test
    void shouldCompleteJobSuccessfully() throws Exception {
        // Given - datos de test
        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 - sin datos

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

        // Then - el job tiene éxito incluso sin datos
        assertThat(execution.getStatus()).isEqualTo(BatchStatus.COMPLETED);
    }

    @Test
    void shouldRestartFromFailurePoint() throws Exception {
        // Given - simula error a mitad de procesamiento
        insertTestOrders(100);
        insertPoisonOrder(50);  // Causa un error

        // When - primera ejecución falla
        JobExecution firstExecution = jobLauncherTestUtils.launchJob();
        assertThat(firstExecution.getStatus()).isEqualTo(BatchStatus.FAILED);

        // Corrige los datos
        removePoisonOrder(50);

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

        // Then - reanuda desde el punto de fallo
        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);
    }
}

// Writer optimizado para grandes volúmenes
@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;
        }

        // Usa PreparedStatement con 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();
            }

            // Ejecuta todos los inserts en una sola operación de red
            ps.executeBatch();
        }
    }
}

// StagingTableWriter.java
// Patrón de tabla de staging para volúmenes muy grandes
@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) {
        // Crea una tabla temporal para este 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) {
        // Escribe en la tabla de staging (sin constraints FK)
        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 a la tabla final
            jdbcTemplate.execute(
                    "INSERT INTO processed_orders SELECT * FROM " + stagingTable);
        }
        // Limpia la tabla de staging
        jdbcTemplate.execute("DROP TABLE IF EXISTS " + stagingTable);
        return stepExecution.getExitStatus();
    }
}