We performed a comparison between Amazon Kinesis and Apache Flink based on our users’ reviews in five categories. After reading all of the collected data, you can find our conclusion below.
Comparison Results: Based on the parameters we compared, users are happier with Amazon Kinesis. Although it is not open-source like Apache Flink, Amazon Kinesis users were more satisfied with how the product performed, Apache Flink users were less satisfied with the overall functionality of the product, including its lack of stability and scalability.
"One of the best features of Amazon Kinesis is the multi-partition."
"Great auto-scaling, auto-sharing, and auto-correction features."
"What turns out to be most valuable is its integration with Lambda functions because you can process the data as it comes in. As soon as data comes, you'll fire a Lambda function to process a trench of data."
"Amazon Kinesis also provides us with plenty of flexibility."
"The feature that I've found most valuable is the replay. That is one of the most valuable in our business. We are business-to-business so replay was an important feature - being able to replay for 24 hours. That's an important feature."
"Setting Amazon Kinesis up is quick and easy; it only takes a few minutes to configure the necessary settings and start using it."
"I have worked in companies that build tools in-house. They face scaling challenges."
"The solution has the capacity to store the data anywhere from one day to a week and provides limitless storage for us."
"The event processing function is the most useful or the most used function. The filter function and the mapping function are also very useful because we have a lot of data to transform. For example, we store a lot of information about a person, and when we want to retrieve this person's details, we need all the details. In the map function, we can actually map all persons based on their age group. That's why the mapping function is very useful. We can really get a lot of events, and then we keep on doing what we need to do."
"The product helps us to create both simple and complex data processing tasks. Over time, it has facilitated integration and navigation across multiple data sources tailored to each client's needs. We use Apache Flink to control our clients' installations."
"Another feature is how Flink handles its radiuses. It has something called the checkpointing concept. You're dealing with billions and billions of requests, so your system is going to fail in large storage systems. Flink handles this by using the concept of checkpointing and savepointing, where they write the aggregated state into some separate storage. So in case of failure, you can basically recall from that state and come back."
"The top feature of Apache Flink is its low latency for fast, real-time data. Another great feature is the real-time indicators and alerts which make a big difference when it comes to data processing and analysis."
"Apache Flink is meant for low latency applications. You take one event opposite if you want to maintain a certain state. When another event comes and you want to associate those events together, in-memory state management was a key feature for us."
"The documentation is very good."
"This is truly a real-time solution."
"It is user-friendly and the reporting is good."
"There are certain shortcomings in the machine learning capacity offered by the product, making it an area where improvements are required."
"Kinesis Data Analytics needs to be improved somewhat. It's SQL based data but it is not as user friendly as MySQL or Athena tools."
"Could include features that make it easier to scale."
"In order to do a successful setup, the person handling the implementation needs to know the solution very well. You can't just come into it blind and with little to no experience."
"We were charged high costs for the solution’s enhanced fan-out feature."
"Kinesis is good for Amazon Cloud but not as suitable for other cloud vendors."
"I think the default settings are far too low."
"Something else to mention is that we use Kinesis with Lambda a lot and the fact that you can only connect one Stream to one Lambda, I find is a limiting factor. I would definitely recommend to remove that constraint."
"One way to improve Flink would be to enhance integration between different ecosystems. For example, there could be more integration with other big data vendors and platforms similar in scope to how Apache Flink works with Cloudera. Apache Flink is a part of the same ecosystem as Cloudera, and for batch processing it's actually very useful but for real-time processing there could be more development with regards to the big data capabilities amongst the various ecosystems out there."
"Amazon's CloudFormation templates don't allow for direct deployment in the private subnet."
"We have a machine learning team that works with Python, but Apache Flink does not have full support for the language."
"The TimeWindow feature is a bit tricky. The timing of the content and the windowing is a bit changed in 1.11. They have introduced watermarks. A watermark is basically associating every data with a timestamp. The timestamp could be anything, and we can provide the timestamp. So, whenever I receive a tweet, I can actually assign a timestamp, like what time did I get that tweet. The watermark helps us to uniquely identify the data. Watermarks are tricky if you use multiple events in the pipeline. For example, you have three resources from different locations, and you want to combine all those inputs and also perform some kind of logic. When you have more than one input screen and you want to collect all the information together, you have to apply TimeWindow all. That means that all the events from the upstream or from the up sources should be in that TimeWindow, and they were coming back. Internally, it is a batch of events that may be getting collected every five minutes or whatever timing is given. Sometimes, the use case for TimeWindow is a bit tricky. It depends on the application as well as on how people have given this TimeWindow. This kind of documentation is not updated. Even the test case documentation is a bit wrong. It doesn't work. Flink has updated the version of Apache Flink, but they have not updated the testing documentation. Therefore, I have to manually understand it. We have also been exploring failure handling. I was looking into changelogs for which they have posted the future plans and what are they going to deliver. We have two concerns regarding this, which have been noted down. I hope in the future that they will provide this functionality. Integration of Apache Flink with other metric services or failure handling data tools needs some kind of update or its in-depth knowledge is required in the documentation. We have a use case where we want to actually analyze or get analytics about how much data we process and how many failures we have. For that, we need to use Tomcat, which is an analytics tool for implementing counters. We can manage reports in the analyzer. This kind of integration is pretty much straightforward. They say that people must be well familiar with all the things before using this type of integration. They have given this complete file, which you can update, but it took some time. There is a learning curve with it, which consumed a lot of time. It is evolving to a newer version, but the documentation is not demonstrating that update. The documentation is not well incorporated. Hopefully, these things will get resolved now that they are implementing it. Failure is another area where it is a bit rigid or not that flexible. We never use this for scaling because complexity is very high in case of a failure. Processing and providing the scaled data back to Apache Flink is a bit challenging. They have this concept of offsetting, which could be simplified."
"The machine learning library is not very flexible."
"In a future release, they could improve on making the error descriptions more clear."
"Apache Flink's documentation should be available in more languages."
"PyFlink is not as fully featured as Python itself, so there are some limitations to what you can do with it."
Amazon Kinesis is ranked 1st in Streaming Analytics with 24 reviews while Apache Flink is ranked 5th in Streaming Analytics with 15 reviews. Amazon Kinesis is rated 8.0, while Apache Flink is rated 7.6. The top reviewer of Amazon Kinesis writes "Used for media streaming and live-streaming data". On the other hand, the top reviewer of Apache Flink writes "A great solution with an intricate system and allows for batch data processing". Amazon Kinesis is most compared with Azure Stream Analytics, Amazon MSK, Confluent, Google Cloud Dataflow and Apache Spark Streaming, whereas Apache Flink is most compared with Spring Cloud Data Flow, Databricks, Azure Stream Analytics, Apache Pulsar and Google Cloud Dataflow. See our Amazon Kinesis vs. Apache Flink report.
See our list of best Streaming Analytics vendors.
We monitor all Streaming Analytics reviews to prevent fraudulent reviews and keep review quality high. We do not post reviews by company employees or direct competitors. We validate each review for authenticity via cross-reference with LinkedIn, and personal follow-up with the reviewer when necessary.