Trouble configuring AWS DataPipeline to use Spot Instances instead of On-Demand Instances

Question

Hi Team,

I have set up an **AWS DataPipeline** to run my EMR jobs on `On-Demand` instances. However, I now want to switch to using `Spot` Instances to reduce costs. I have configured the `spotBidPrice` parameter in my pipeline settings, expecting it to run on Spot Instances. However, it seems that the pipeline is still using On-Demand instances.

Could you please help me understand how I can properly configure my DataPipeline to utilize Spot Instances? Here are my current pipeline settings:

```
coreInstanceCount: 40
coreInstanceType: r5.4xlarge
keyPair: #{myKeyPair}
masterInstanceType: r5.4xlarge
maximumRetries: 2
region: #{myRegion}
releaseLabel: emr-5.31.0
resourceRole: #{myResourceRole}
role: #{myRole}
subnetId: #{mySubnetId}
taskInstanceType: r5.4xlarge
terminateAfter: 240 Minutes
spotBidPrice: 10.00
useOnDemandOnLastAttempt: true
```

I appreciate any guidance or suggestions you can provide to help me successfully configure my AWS DataPipeline to use Spot Instances. Thank you!

Answer

Hi,

I can see from the current pipeline settings, that the “useOnDemandOnLastAttempt” is set to “true”. The parameter 'useOnDemandOnLastAttempt' is set to true by default. To avoid getting on demand instances used for EMR cluster, when Spot instances are not available you need to set this parameter to false. Also the maximum attempts for EMR cluster resource is defaulted to "1", you can also change the "maximumRetries" of EMR cluster to more than “1”. Currently you have "maximumRetries: 2” you can increase it to get the spot instances in other attempts.

Scenarios where the spot instances fail to launch

1. Spot price is low than the minimum required Spot request fulfillment price.
2. Limitation issue "EXCEEDED_SPOT_INSTANCE_COUNT_LIMIT (USER_ERROR)".

Answer

Hi there!

Can you try using `taskInstanceBidPrice` instead of `spotBidPrice`?

I hope this helps.

Trouble configuring AWS DataPipeline to use Spot Instances instead of On-Demand Instances

Contenido relevante