Développez l'apprentissage automatique prédictif avec Flink | Atelier du 18 déc. | S'inscrire
Also known as event stream processing (ESP), event streaming patterns can process a continuous flow of data as soon as an event or change happens. By processing single points of data rather than an entire batch, event streaming platforms provide an architecture that enables software to understand, react to, and operate as events occur.
Learn how event stream processing works, its major benefits, and how to get started building event-driven architectures in the free stream processing guide.
An Event Stream is an ordered sequence of events representing important actions in a software domain. This can be something simple, like clicking on a link, or it might be something more complex, like transferring funds between two banks.
Event Stream Processing (ESP) takes a continuous stream of events and processes them as soon as a change happens. By processing single points of data rather than an entire batch, event streaming platforms provide an architecture that enable software to understand, react to, and operate as events occur.
In software, any significant action can be recorded as an event. For example, it could be as simple as someone clicking a link or viewing a webpage, or something more involved like paying for an order, withdrawing money, or even communicating with numerous, distributed IoT devices at once.
These events can be organized into streams, essentially a series of events ordered by time. From there, events can be shared with other systems where they can be processed in real-time. Events are pushed and handled one at a time, as they happen. This allows the system to react in real-time, rather than waiting for batches to accumulate.
For example, each time someone clicks a link or views a webpage, we might push an event into a system such as Apache Kafka. Downstream, a Flink job could consume those events to develop analytics about how many views and clicks our website is receiving.
The most obvious benefit of event streaming is that it allows systems to react in real time. For example, multi-player video games, a stock transaction, or personalized shopping recommendations. This gives users the kind of instant gratification that they crave while better reflecting the world we live in.
However, there are other benefits to Event Streaming, beyond the user experience.
Systems built with event streams are better equipped to react to changes in the environment. If there is a sudden increase in load, these systems can react immediately, rather than having to wait some predefined interval. This allows systems to be more elastic, scaling up and down as the requirements demand.
Event streaming systems also tend to be more resilient. Batch systems are notorious for expensive failures. Because these systems are building up large amounts of data to be processed all at once, a single piece of bad data can collapse the whole process. This can result in large and expensive retries. However, with streaming systems, we deal with the events as they happen. If an event fails for some reason, then we can put processes in place to handle that failure with a minimal amount of interruption to the rest of the events.
Imagine if you walked into the grocery store and were told that you’d have to wait to do your shopping because the store required batches of at least 10 shoppers before they could be allowed in. Or perhaps, you step up to a bank machine to withdraw $20, only to discover that the bank will only release funds in batches of $5000 at a time. On the surface, these examples, sound ridiculous, but the reality is that many software applications are built with this mentality.
Batch processing is a tried and tested method of handling large amounts of data. The system collects data into larger batches and processes it all at once. The approach can be used to optimize applications and can help make a system more efficient. Unfortunately, it also introduces significant latency and rarely reflects how things work in the real world. As users have become more accustomed to real-time experiences, batch processing is unable to keep up.
The real world exists as a continuous stream of events and the reactions to those events. When we walk into a grocery store, that’s an event. Attempting to withdraw $20 from the bank is also an event. When we initiate each of these, there is an expectation that the people and processes we interact with will respond in real-time. We don’t want to wait for some seemingly arbitrary threshold to be crossed. We want results now.
In software, if we want to provide users with real-time experiences, we need to move beyond batch processing and instead adopt event streaming.
Les cas d'usage du streaming événementiel sont nombreux. Parce qu'il se rapproche de la façon dont les choses fonctionnent dans le monde réel, presque tous les processus métier sont mieux représentés avec le streaming événementiel qu'ils ne le seraient avec les traitements par lots. Un avantage qui est particulièrement évident dans certains domaines tels que l'analyse prédictive, le machine learning, l'IA générative, la détection de la fraude, et bien plus encore.
Le streaming événementiel est utilisé dans un grand nombre d'applications, notamment les médias en continu, le retail omnicanal, le covoiturage, etc.
Par exemple, lorsqu'un usager fait appel à Lyft, l'application sait non seulement quel chauffeur lui proposer, mais aussi combien de temps la course prendra grâce à la localisation et à l'historique des données de trafic. Elle peut également déterminer le coût de la course en fonction des données en temps réel et historiques.
– Détection de la fraude – Trading d'actions en temps réel – Marketing, commerce et analyses des données en entreprise – Activité des clients/utilisateurs – Surveillance des systèmes informatiques internes et élaboration de rapports – Suivi de logs : dépannage des systèmes, des serveurs, des appareils, etc. – SIEM (gestion des événements et des informations de sécurité) : analyse en temps réel des fichiers logs et des données événementielles à des fins de suivi, de mesure et de détection des menaces – Gestion des stocks de la vente au détail/des entrepôts : gestion des stocks sur tous les canaux et pour tous les emplacements, et fourniture d'une expérience utilisateur fluide d'un appareil à l'autre – Mise en correspondance des VTC : combiner les données relative à la localisation, à l'utilisateur et au tarif à des fins d'analyse prédictive – Associer les usagers aux meilleurs chauffeurs (selon des critères comme la proximité, la destination, le tarif et le délai d'attente) – Apprentissage automatique et IA : la combinaison de données historiques et actuelles au sein d'un système centralisé ouvre de nouvelles perspectives pour l'analyse prédictive – Analyse prédictive
In the world of event streaming, Kafka and Flink are two of the most recognizable technologies in use today.
Apache Kafka is where your event streams live. Each event gets pushed into a Kafka topic where it can be consumed by any downstream systems interested in reacting to those events.
Apache Flink is a powerful engine for processing events in real time. It provides a robust set of features to consume the events and perform analytics against them.
Together, event streaming with Kafka and Flink act as a central nervous system capable of reacting to millions of events.
Built by the original creators of Apache Kafka, Confluent takes Kafka's stream processing technology to a fully managed, multi-cloud data streaming platform. Easily connect 120+ data sources with enterprise-grade security, performance, and scalability. Stream data across any infrastructure in minutes.
More recently, Confluent has introduced Apache Flink into the cloud platform to provide a first-class stream processing engine, alongside its already robust streaming system.
Used by 80% of the Fortune 100, Confluent's data streaming platform helps you set your data in motion, no matter where it resides.