Senior Technical Engineer at a transportation company with 5,001-10,000 employees
Real User
Top 10
Dec 26, 2025
I feel some lack of functionality in Amazon EMR. I have thoughts on what would be great to see in the product, such as AI/ML features or additional options.
Senior Chief Engineer (Enterprise System Presales/Postsales) at a tech vendor with 10,001+ employees
Real User
Top 20
Oct 6, 2025
I have used AWS Glue with S3 for making tables and databases, but regarding Amazon EMR, I do not remember much as we are currently using it very minimally. This is my observation: In EKS, we have had to deploy by ourselves because EKS does not provide the Hadoop framework, Spark, Hive, and everything, but we have completed all the deployment ourselves. Whereas Amazon EMR provides all these things. The cost factor differs significantly. When you run Spark application on EKS, you run at the pod level, so you can control the compute cost. But in Amazon EMR, when you have to run one application, you have to launch the entire EC2. In Qubole, the interface was very good. I could see many details because in Amazon EMR console, very few details are available. In Qubole, at one link, you can get all the details of what is happening, how the processes are running, and the cost decreased by using Qubole. I found Qubole more user-friendly and cost-effective. From the security point of view, we had to open some access rights to Qubole, which might be a drawback in comparison to Amazon EMR which is native to AWS.
There is room for improvement with respect to retries, handling the volume of data on S3 ( /products/amazon-s3-reviews ) buckets, cluster provisioning, scaling, termination, security, and integration between services like S3, Glue, Lake Formation, and DynamoDB.
As people are shifting from legacy solutions to other technologies, Amazon EMR needs to add more features that give more flexibility in managing user data.
Lead Data Engineer at Seven Lakes Enterprises, Inc.
Real User
Aug 15, 2023
Interdependencies with a third-party or open source solution should be improved. Modules and strategies should be better handled and notified early in advance. Maybe if AWS starts releasing AWS-certified or AWS-verified installations, that will generate even more confidence just like OpenJet, it'll add a specific version.
Technology Analyst at Proodlehospitatilityservicesltd
Real User
Jul 27, 2023
The product's features for storing data in static clusters could be better. It would be helpful if they released a beta version for limited users to know about the product.
We have had issues with the boolean mathematical operation in 2. X's big version is working in newer versions because the old version of the solution does not support it, which is a compatibility issue that can be improved. In addition, the legacy versions of the solution are not supported in the new versions.
The problem for us is it starts very slow. They need to improve the start time. If we use a long-running EMR, it costs a lot of money. However, when we start, for example, a job, if the job runs for one hour, it's normal as it starts in about ten minutes. If we want, for example, to run each five minutes, it's a problem if it takes ten minutes to start. It's a little bit weird that you cannot use the service within a short period. The support could be better.
The cost is increasing. We are looking into how we can optimize the cost part of EMR. We're doing a comparison between Cloudera running on AWS and running AWS EMR. We don't have much control. If we have multiple users, if they want to scale up, the cost will go and increase and we don't know how we can restrict that price part.
Deputy CTO at a tech company with 51-200 employees
Real User
Jun 25, 2021
The most complicated thing is configuring to the cluster and to ensure it's running correctly. You need to configure at least three Amazon policies to get authorization for all the instances. And if you're new on the system it's really complicated. It's something that could be simplified for users. For additional features, I'd like to see a better MLOps platform but it's possible that it's already in production.
The dashboard management could be better. Right now, it's lacking a bit. I'd like more of a remote connection between my computer and the solution. We have multi-factor authentication, and at one point it was an issue due to the fact that I lost my phone. It stopped me from accessing the system. We have to replicate all the infrastructure and we need to ensure that we have the scalability and to do so in production. We are hoping that Amazon will allow us to scale easily. However, we have not attempted to scale just yet.
Amazon EMR simplifies big data processing by offering integration with popular tools. It's scalable and cost-efficient, enabling fast processing while managing infrastructure effortlessly. It's designed for users aiming to streamline data workflows and leverage its batch processing capabilities effectively.Amazon EMR is a managed service that provides robust features for big data processing. It integrates seamlessly with S3, EC2, Hive, and Spark to facilitate sophisticated data transformation...
I feel some lack of functionality in Amazon EMR. I have thoughts on what would be great to see in the product, such as AI/ML features or additional options.
I have used AWS Glue with S3 for making tables and databases, but regarding Amazon EMR, I do not remember much as we are currently using it very minimally. This is my observation: In EKS, we have had to deploy by ourselves because EKS does not provide the Hadoop framework, Spark, Hive, and everything, but we have completed all the deployment ourselves. Whereas Amazon EMR provides all these things. The cost factor differs significantly. When you run Spark application on EKS, you run at the pod level, so you can control the compute cost. But in Amazon EMR, when you have to run one application, you have to launch the entire EC2. In Qubole, the interface was very good. I could see many details because in Amazon EMR console, very few details are available. In Qubole, at one link, you can get all the details of what is happening, how the processes are running, and the cost decreased by using Qubole. I found Qubole more user-friendly and cost-effective. From the security point of view, we had to open some access rights to Qubole, which might be a drawback in comparison to Amazon EMR which is native to AWS.
There is room for improvement with respect to retries, handling the volume of data on S3 ( /products/amazon-s3-reviews ) buckets, cluster provisioning, scaling, termination, security, and integration between services like S3, Glue, Lake Formation, and DynamoDB.
Spark jobs take longer on Amazon EMR compared to previous experiences. This aspect could be improved to make them more efficient.
The solution can become expensive if you are not careful.
As people are shifting from legacy solutions to other technologies, Amazon EMR needs to add more features that give more flexibility in managing user data.
The product's stability could be even better.
Interdependencies with a third-party or open source solution should be improved. Modules and strategies should be better handled and notified early in advance. Maybe if AWS starts releasing AWS-certified or AWS-verified installations, that will generate even more confidence just like OpenJet, it'll add a specific version.
The product's features for storing data in static clusters could be better. It would be helpful if they released a beta version for limited users to know about the product.
We have had issues with the boolean mathematical operation in 2. X's big version is working in newer versions because the old version of the solution does not support it, which is a compatibility issue that can be improved. In addition, the legacy versions of the solution are not supported in the new versions.
The product can be improved by automating their up-sizing and downsizing their cluster.
The problem for us is it starts very slow. They need to improve the start time. If we use a long-running EMR, it costs a lot of money. However, when we start, for example, a job, if the job runs for one hour, it's normal as it starts in about ten minutes. If we want, for example, to run each five minutes, it's a problem if it takes ten minutes to start. It's a little bit weird that you cannot use the service within a short period. The support could be better.
The cost is increasing. We are looking into how we can optimize the cost part of EMR. We're doing a comparison between Cloudera running on AWS and running AWS EMR. We don't have much control. If we have multiple users, if they want to scale up, the cost will go and increase and we don't know how we can restrict that price part.
Amazon EMR is continuously improving, but maybe something like CI/CD out-of-the-box or integration with Prometheus Grafana.
The most complicated thing is configuring to the cluster and to ensure it's running correctly. You need to configure at least three Amazon policies to get authorization for all the instances. And if you're new on the system it's really complicated. It's something that could be simplified for users. For additional features, I'd like to see a better MLOps platform but it's possible that it's already in production.
The dashboard management could be better. Right now, it's lacking a bit. I'd like more of a remote connection between my computer and the solution. We have multi-factor authentication, and at one point it was an issue due to the fact that I lost my phone. It stopped me from accessing the system. We have to replicate all the infrastructure and we need to ensure that we have the scalability and to do so in production. We are hoping that Amazon will allow us to scale easily. However, we have not attempted to scale just yet.