How to ensure EMR process steps in sequence?

Question

I'm submitting multiple steps with AWS python SDK

```py
steps = []
for job in jobs:
    args = [
        'spark-submit',
        '--py-files',
        's3://bucket/scripts/*',
        's3://bucket/scripts/main.py',
    ]
    args = args + job.params
    step = {
        'Name': job.name,
        'ActionOnFailure': 'CONTINUE',
        'HadoopJarStep': {
            'Jar': 'command-runner.jar',
            'Args': args
        }
    }
    steps.append(step)
response = self.client.add_job_flow_steps(JobFlowId=job_flow_id, Steps=steps)
```

But EMR does not process the step in the sequence as the `steps` array, is there a way to ensure it processes in sequence?

Answer

I believe that steps are submitted and run in the order, so to confirm the same I went ahead and tested the same on a test cluster using your code with little changes as below.

```
steps = []
for i in range(1,10):
    args = [
        'spark-example',
        '--deploy-mode',
        'cluster',
        'SparkPi',
        '10'
    ]
    step = {
        'Name': "TestStepOrder" + str(i),
        'ActionOnFailure': 'CONTINUE',
        'HadoopJarStep': {
            'Jar': 'command-runner.jar',
            'Args': args
        }
    }
    steps.append(step)
response = client.add_job_flow_steps(JobFlowId=clusterId, Steps=steps)

```

I can confirm the order is maintained as expected.

I ran a second round of tests for the same with concurrency set as 5 to see if that has any impacts on this. In this case by looking at the Start Time I can confirm the order is still maintained.

Interested to know more about how you get the order mixed up, please share reproduction steps to reproduce the behavior you are observing.

*Note: I'm using the latest boto3 version(1.20.26), not sure if that makes it any different*

How to ensure EMR process steps in sequence?

相关内容