Conditional (Java configured) steps cannot be scoped [BATCH-2747] #857
Comments
|
Michael Minella commented What is the use case where the beans must be scoped? |
|
Mattias Jiderhamn commented Michael Minella, the most obvious use case would be when JobParameters are used in the step definition @Bean
@JobScope
Step readDataFromFile(@Value("#{jobParameters[filePath]}") String filePath) {
...
}
There does however seem to be at least one workaround, by separately defining the @Bean
@JobScope
ItemReader reader(@Value("#{jobParameters[filePath]}") String filePath) {
...
}
@Bean(name = "stepA")
Step readDataFromFile(ItemReader reader) {
...
}
@Bean
Job myJob(@Qualifier("stepA") stepA) {
return this.jobBuilderFactory.get("job")
.start(stepA)
.on("*").to(stepB())
.from(stepA).on("FAILED").to(stepC())
.end()
.build();
} |
|
ArtyomGabeev commented Hi, I think issue is related to FlowBuilder - it's impossible to use scoped steps/flows inside any other flow. Here is a small example: @Configuration
public class JobWithFlowConfig {
@Bean
public Job job(JobBuilderFactory jobBuilderFactory, Flow flow) {
return jobBuilderFactory.get("job")
.start(flow)
.end()
.build();
}
@Bean
public Flow flow(Step step) {
return new FlowBuilder<Flow>("flow")
.from(step)
.end();
}
@Bean
@JobScope
public Step step(StepBuilderFactory stepBuilderFactory, @Value("#{jobParameters['test']}") String test) {
return stepBuilderFactory.get("step")
.tasklet(((contribution, chunkContext) -> {
System.out.println(test);
return RepeatStatus.FINISHED;
}))
.allowStartIfComplete(true)
.build();
}
}
Issue happens in FlowBuilder.createState - calling getName on a step/flow causes lazy proxy initialization, but there is no Job/Step scope at this time. In order to fix it, I suggest reuse logic of conters naming for step/flow: ...
private int stepCounter = 0;
private int flowCounter = 0;
...
private State createState(Object input) {
...
states.put(input, new StepState(prefix + "step" + (stepCounter++), step)); //remove getName call
...
states.put(input, new FlowState((Flow) input, prefix + "flow" + (flowCounter++))); //same for the flow
}Thanks, |
|
ArtyomGabeev commented Regarding original issue: calling on over jobBuilder, simply converts previous steps into FlowBuilder, that explains the behaviour. |
|
Michael Minella commented This still doesn't make sense to me. You need to be in the context of a step in order to to use step scope. What you are trying to do is causing premature initialization. I don't see this as a bug but works as designed. |
|
ArtyomGabeev commented What about JobScope? In general step can be annotated with
Michael Minella, you right, it was my mistake when I mentioned StepScope in this issue, because it doesn't make sense. But this fix is still reasonable if you want to create a JobScoped step, and pass some parameters (like chunk size for example)
BTW: original author has an issue with JobScoped step, according to stack trace. |
|
Michael Minella commented This really also works as designed. There must be a running job in order for the proxy to be resolved. The steps are created before the job is executed so the idea that you can use job scope on a step still doesn't apply. I'd argue that we may need to be more specific around our documentation, but you do not have a running job when a step is configured so there is no way for job scope to work on the configuration of a step as you are attempting to do. |
|
ArtyomGabeev commented Thanks, now I see your point. My initial understanding was that JobScope can be applied on anything except Job. So there is no way to bind jobParameters to step configuration, so let's say I can not adjust step chunk size based on jobParameters? Another use case is to configure gridSize based on jobParameters. |
|
ArtyomGabeev commented Also, it's possible to configure JobScope'd step, if there is no flow: @Configuration
public class JobConfig {
@Bean
public Job job(JobBuilderFactory jobBuilderFactory, Step step) {
return jobBuilderFactory.get("job")
.start(step)
.build();
}
@Bean
@JobScope
public Step step(StepBuilderFactory stepBuilderFactory, @Value("#{jobParameters['test']}") String test) {
return stepBuilderFactory.get("step")
.tasklet(((contribution, chunkContext) -> {
System.out.println(test);
return RepeatStatus.FINISHED;
}))
.allowStartIfComplete(true)
.build();
}
} This configuration works fine. Also I want to try same two configurations in XML to confirm behaviour. |
|
ArtyomGabeev commented Finally, I agree with that step can not be scoped, when I get to the partitioned steps (there is no job scope). Probably documentation should be more specific in terms what can be proxied. Thank you for clarification.
|
|
Mattias Jiderhamn commented Sorry for being late back to the party. Even though it may "work as designed", at least personally I think that design is very far from intuitive and definitely could use clarifying in the documentation. I spent quite some time figuring out a working configuration using trial and error.
+Despite having read the documentation+ this really doesn't make sense to me. If you create a job scoped or step scoped bean - such as a reader, as in my example from September 1st - you may use that bean when configuring your step. But you may not make the step itself job scoped nor step scoped - am I right? This seems like saying steps may be implicitly job scoped or step scoped (due to the dependency on the scoped bean), but not explicitly. That is - with the current documentation - counter-intuitive. I mean, what are the job and step scopes for, if not for being used in the configuration???
Here is a variation of my September 1st example. The fact that it also works for me only adds to the confusion. @Bean
@JobScope
ItemReader reader(@Value("#{jobParameters[filePath]}") String filePath) {
...
}
@Bean
Step readDataFromFile(ItemReader reader) {
return steps.get("readDataFromFileStep")
.<X, Y>chunk(100)
.reader(reader("Overridden by expression"))
...
}(Sorry for spamming you with updates fixing Jira copy/paste issues) |
spring-projects-issues
added
type: bug
status: waiting-for-triage
|
My apologies for asking what may be a dumb question. I hit this issue today by using a combination of Java-configured conditional flow with some steps that have I'm not quite understanding the current conclusion of the above thread - is that something that "should work", or is it "by design"? If the latter, is there a recommended workaround if I need my |
|
@wabrit Please ask these kind of questions via StackOverflow and tag them with the |
mminella
added
in: documentation
type: enhancement
and removed
in: core
status: waiting-for-triage
Thanks @mminella - regardless of the status/scope of this particular issue (which I found unclear from reading the trail above) if it's of any use to you I can confirm that:
|
|
@wabrit Those are as expected. We will be updating the documentation to make that more clear. Thanks! |
Mattias Jiderhamn opened BATCH-2747 and commented
Configuring conditional steps using Java config (XML untested) fails if any of the steps is scoped using Job scope or Step scope.
To reproduce, use the example at https://docs.spring.io/spring-batch/4.0.x/reference/html/step.html#controllingStepFlow, i.e.
and then add
@JobScopeor@StepScopeonstepA(),stepB()orstepC().The underlying reason is that to configure a flow step, the framework wants to get the steps name using
getName(), at which point the bean is resolved and the scope inferred.Stack trace (4.0.1)
Affects: 4.0.1, 4.1.0.M2
Reference URL: https://stackoverflow.com/questions/55571210/
2 votes, 4 watchers
The text was updated successfully, but these errors were encountered: