SPRING BATCH中的REMOTE CHUNKING JOB，由于是基于MASTER/SLAVE的架构，其中某个STEP是会在远程机器中执行，如果要停止这个JOB，需要考虑两个问题：
1、什么时候发出停止指令
2、如何等待远程STEP的完成
一般停止JOB，可用JobOperator.stop(long executionId)来停止，但这个无法确定什么时候发出停止指令，如果是在CHUNK的处理中途发出，则会出现回滚的现象。
BATCH_STEP_EXECUTION
thead tr {background-color: ActiveCaption; color: CaptionText;}
th, td {vertical-align: top; font-family: "Tahoma", Arial, Helvetica, sans-serif; font-size: 8pt; padding: 4px; }
table, td {border: 1px solid silver;}
table {border-collapse: collapse;}
thead .col0 {width: 173px;}
.col0 {text-align: right;}
thead .col1 {width: 82px;}
.col1 {text-align: right;}
thead .col2 {width: 282px;}
thead .col3 {width: 164px;}
.col3 {text-align: right;}
thead .col4 {width: 161px;}
thead .col5 {width: 161px;}
thead .col6 {width: 109px;}
thead .col7 {width: 127px;}
.col7 {text-align: right;}
thead .col8 {width: 109px;}
.col8 {text-align: right;}
thead .col9 {width: 118px;}
.col9 {text-align: right;}
thead .col10 {width: 117px;}
.col10 {text-align: right;}
thead .col11 {width: 142px;}
.col11 {text-align: right;}
thead .col12 {width: 150px;}
.col12 {text-align: right;}
thead .col13 {width: 166px;}
.col13 {text-align: right;}
thead .col14 {width: 137px;}
.col14 {text-align: right;}
thead .col15 {width: 109px;}
thead .col16 {width: 156px;}
thead .col17 {width: 161px;}
| STEP_EXECUTION_ID |
VERSION |
STEP_NAME |
JOB_EXECUTION_ID |
START_TIME |
END_TIME |
STATUS |
COMMIT_COUNT |
READ_COUNT |
FILTER_COUNT |
WRITE_COUNT |
READ_SKIP_COUNT |
WRITE_SKIP_COUNT |
PROCESS_SKIP_COUNT |
ROLLBACK_COUNT |
EXIT_CODE |
EXIT_MESSAGE |
LAST_UPDATED |
| 2304 |
169 |
step2HandleXXX |
434 |
2020-06-22 16:27:54 |
2020-06-22 16:32:46 |
STOPPED |
167 |
5010 |
0 |
4831 |
0 |
155 |
0 |
161 |
STOPPED |
org.springframework.batch.core.JobInterruptedException |
2020-06-22 16:32:46 |
另外SPRING BATCH也不会等远程STEP执行完成，就将JOB的状态设为Complete。
发出停止的指令应通过ChunkListener达成：
public class ItemMasterChunkListener extends ChunkListenerSupport{
private static final Logger log = LoggerFactory.getLogger(ItemMasterChunkListener.class);
@Override
public void beforeChunk(ChunkContext context) {
log.info("ItemMasterProcessor.beforeChunk");
}
@Override
public void afterChunk(ChunkContext context) {
log.info("ItemMasterProcessor.afterChunk");
if(XXXX.isStoppingOrPausing()) {
log.info("context.getStepContext().getStepExecution().setTerminateOnly()");
context.getStepContext().getStepExecution().setTerminateOnly();
}
}
@Override
public void afterChunkError(ChunkContext context) {
log.info("ItemMasterProcessor.afterChunkError");
}
}
配置BEAN：
@Bean
@StepScope
public ItemMasterChunkListener novaXItemMasterChunkListener() {
return new ItemMasterChunkListener();
}
this.masterStepBuilderFactory
.<X, X>get("step2Handle")
.listener(itemMasterChunkListener())
.build();
由于是在CHUNK完成的时候发出停止指令，就不会出现ROLLBACK的情况。
等待远程STEP完成，通过读取MQ上的MESSAGE是否被消费完成，PENDDING的MESSAGE为0的条件即可。
public class JobExecutionListenerSupport implements JobExecutionListener {
/* (non-Javadoc)
* @see org.springframework.batch.core.domain.JobListener#afterJob()
*/
@Override
public void afterJob(JobExecution jobExecution) {
Integer totalPendingMessages = 0;
String queueName = "";
String messageSelector = "JOB_EXECUTION_ID=" + jobExecution.getJobInstance().getInstanceId();
do{
totalPendingMessages =
this.jmsTemplate.browseSelected(queueName, messageSelector,
(session, browser) ->
Collections.list(browser.getEnumeration()).size()
);
String brokerURL = null;
if(jmsTemplate.getConnectionFactory() instanceof JmsPoolConnectionFactory) {
JmsPoolConnectionFactory connectionFactory =
(JmsPoolConnectionFactory)jmsTemplate.getConnectionFactory();
ActiveMQConnectionFactory activeMQConnectionFactory =
(ActiveMQConnectionFactory)connectionFactory.getConnectionFactory();
brokerURL = activeMQConnectionFactory.getBrokerURL();
} else if(jmsTemplate.getConnectionFactory() instanceof CachingConnectionFactory) {
CachingConnectionFactory connectionFactory =
(CachingConnectionFactory)jmsTemplate.getConnectionFactory();
ActiveMQConnectionFactory activeMQConnectionFactory =
(ActiveMQConnectionFactory)connectionFactory.getTargetConnectionFactory();
brokerURL = activeMQConnectionFactory.getBrokerURL();
}
LOGGER.info("queueName = {}, {}, totalPendingMessages = {}, url={}",
queueName, messageSelector, totalPendingMessages, brokerURL);
Assert.notNull(totalPendingMessages, "totalPendingMessages must not be null.");
try {
Thread.sleep(5_000);
} catch (InterruptedException e) {
LOGGER.error(e.getMessage(), e);
}
} while(totalPendingMessages.intValue() > 0);
}
/* (non-Javadoc)
* @see org.springframework.batch.core.domain.JobListener#beforeJob(org.springframework.batch.core.domain.JobExecution)
*/
@Override
public void beforeJob(JobExecution jobExecution) {
}
}
这样整个JOB就能无异常地停止，且会等待远程STEP完成。
Reference:
https://docs.spring.io/spring-batch/docs/4.1.3.RELEASE/reference/html/common-patterns.html#stoppingAJobManuallyForBusinessReasons
https://stackoverflow.com/questions/13603949/count-number-of-messages-in-a-jms-queue
https://stackoverflow.com/questions/55499965/spring-batch-stop-job-execution-from-external-class
https://stackoverflow.com/questions/34621885/spring-batch-pollable-channel-with-replies-contains-chunkresponses-even-if-job