Apache Flink vs Apache Spark Streaming comparison

"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."

"Apache Flink's best feature is its data streaming tool."

"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 allows you to reduce latency and process data in real-time, making it ideal for such scenarios."

"Apache Flink offers a range of powerful configurations and experiences for development teams. Its strength lies in its development experience and capabilities."

"It is user-friendly and the reporting is good."

"Allows us to process batch data, stream to real-time and build pipelines."

"The setup was not too difficult."

"The solution is better than average and some of the valuable features include efficiency and stability."

"As an open-source solution, using it is basically free."

"Apache Spark Streaming was straightforward in terms of maintenance. It was actively developed, and migrating from an older to a newer version was quite simple."

"Apache Spark's capabilities for machine learning are quite extensive and can be used in a low-code way."

"Apache Spark Streaming is versatile. You can use it for competitive intelligence, gathering data from competitors, or for internal tasks like monitoring workflows."

"The solution is very stable and reliable."

"Apache Spark Streaming's most valuable feature is near real-time analytics. The developers can build APIs easily for a code-steaming pipeline. The solutions have an ecosystem of integration with other stock services."

"Apache Spark Streaming has features like checkpointing and Streaming API that are useful."

"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."

"In a future release, they could improve on making the error descriptions more clear."

"In terms of improvement, there should be better reporting. You can integrate with reporting solutions but Flink doesn't offer it themselves."

"In terms of stability with Flink, it is something that you have to deal with every time. Stability is the number one problem that we have seen with Flink, and it really depends on the kind of problem that you're trying to solve."

"The machine learning library is not very flexible."

"The state maintains checkpoints and they use RocksDB or S3. They are good but sometimes the performance is affected when you use RocksDB for checkpointing."

"We have a machine learning team that works with Python, but Apache Flink does not have full support for the language."

"There are more libraries that are missing and also maybe more capabilities for machine learning."

"The solution itself could be easier to use."

"It was resource-intensive, even for small-scale applications."

"Integrating event-level streaming capabilities could be beneficial."

"In terms of improvement, the UI could be better."

"The initial setup is quite complex."

"The debugging aspect could use some improvement."

"We would like to have the ability to do arbitrary stateful functions in Python."

"The cost and load-related optimizations are areas where the tool lacks and needs improvement."

"The solution is open-source, which is free."

"It's an open source."

"This is an open-source platform that can be used free of charge."

"Apache Flink is open source so we pay no licensing for the use of the software."

"It's an open-source solution."

"Spark is an affordable solution, especially considering its open-source nature."

"People pay for Apache Spark Streaming as a service."

"On a scale from one to ten, where one is expensive, or not cost-effective, and ten is cheap, I rate the price a seven."

"I was using the open-source community version, which was self-hosted."