Spring batch: Fault Tolerant

Question

I have the following Step:

    return stepBuilderFactory.get("billStep")
            .allowStartIfComplete(true)
            .chunk(20000)
            .reader(billReader)
            .processor(billProcessor)
            .faultTolerant()
            .skipLimit(Integer.MAX_VALUE)
            .skip(BillSkipException.class)
            .listener(billReaderListener)
            .listener(billSkipListener)
            .writer(billRepoItemWriter)
            .build();

Is my understanding correct, that fault tolerant means that when an exception is thrown in billProcessor, it will be processed in skip listener and then the next row/item will be processed in billProcessor?

I noticed upon adding in debug logs - that items/rows were "re-processed" when an exception is thrown in the processor. (probably because of faultTolerant config. But, what if I am processing 2 million records, and 300,000 of them were skipped - or throws a skip exception - isn't it an issue in performance if some of these were "re-processed")

And the big problem is - the next row/item is skipped. They were not processed in the processor at all.

If I remove the faultTolerant and SkipListener - and directly save the skipped records in the database (what skiplistener is doing) - it is working, but is this solution correct?

Spring Batch will first process the whole chunk at once (in your case 2000) if that fails it will fallback to proces each item individually so it is able to determine which items where faulty and skipped. Will it impact performance yes, but the whole point of batch processing is not performance but stability. You can either proces nothing or proces what is being able to proces. Then fix the wrong records and run the job again. (The latter is what is possible with springbatch). — M. Deinum, May 30 '17 at 06:46

score 3 · Answer 1 · edited Apr 23 '21 at 17:15

No job is perfect! Errors happen. You may receive bad data. You may forget one null check that causes a NullPointerException at the worst of times. How you handle errors using Spring Batch is our topic today. There are many scenarios where exceptions encountered while processing should not result in Step failure, but should be skipped instead.

Spring batch Skip technique With the skip technique you may specify certain exception types and a maximum number of skipped items, and whenever one of those skippable exceptions is thrown, the batch job doesn’t fail but skip the item and goes on with the next one. Only when the maximum number of skipped items is reached, the batch job will fail. For example, Spring Batch provides the ability to skip a record when a specified Exception is throw when there is an error reading a record from your input. This section will look at how to use this technique to skip records based upon specific Exceptions. There are two pieces involved in choosing when a record is skipped.

1. Exception Under what conditions to skip the record, specifically what exceptions you will ignore. When any error occurs during the reading process, Spring Batch throws an exception. In order to determine what to skip, you need to identify what exceptions to skip.

2. Skipped records How many input records you will allow the step to skip before considering the step execution failed. If you skip one or two records out of a million, not a big deal; however, skipping half a million out of a million is probably wrong. It’s your responsibility to determine the threshold.

(Spring Batch Exception Handling Example)

This entire processing happens at each individual item level, and not a chunk level. Hence, whenever we whenever the spring batch is not able to process an item at a single go. It tries to re-process/dig down into individual items to determine the exact item to skip. This is fine because with batch jobs we expect certain latency as they usually deal with scheduled big data jobs.

score 2 · Answer 2 · answered Jun 11 '20 at 09:46

I noticed upon adding in debug logs - that items/rows were "re-processed" when an exception is thrown in the processor. (probably because of faultTolerant config. But, what if I am processing 2 million records, and 300,000 of them were skipped - or throws a skip exception - isn't it an issue in performance if some of these were "re-processed")

I occured same issue and I fixed using processorNonTransactional method.

    @Bean
    public Step myStep() {
        return stepBuilderFactory.get("myStep")
                .<MyObject, MyObject>chunk(1000)
                .reader(myItemReader())
                .processor(myItemProcessor())
                .writer(jdbcBatchItemWriter())
                .faultTolerant()
                .processorNonTransactional()
                .skip(MyException.class)
                .skipLimit(200)
                .build();
    }

"Mark the item processor as non-transactional (default is the opposite). If this flag is set the results of item processing are cached across transactions in between retries and during skip processing, otherwise the processor will be called in every transaction." https://docs.spring.io/spring-batch/docs/current/api/org/springframework/batch/core/step/builder/FaultTolerantStepBuilder.html#processorNonTransactional-- — bmaupin, Apr 23 '21 at 20:55

Spring batch: Fault Tolerant

2 Answers2