Open broadcast of the first day of the HighLoad ++ conference
On November 7, the reports will begin broadcasting at 9:45, at 18:00 they will end, and the HighLoad ++ Award ceremony will begin.
Open broadcast of the second day of the HighLoad ++ conference
November 8, the first report in the Main Hall and the broadcast begins at 11:00.
Under a cat the schedule and information on reports.
November 7th
10:00 - Why do you need an interservice interaction platform and how to build it today?
Artemy Ryabinkov (Avito)We hope you have already started moving from a monolith to microservices. Then it's time to think about the interservice interaction platform. Indeed, in service architecture, the complexity of support grows exponentially with the number of services. Artemy Ryabinkov promised to give a tool that will significantly slow down the growth of this complexity. His report will be of interest both to those who are just starting their journey in SOA, and to those who are already familiar with the tools OpenAPI, Swagger, gRPC, protobuf.
11:00 - Kafka. Description of one fight
Denis Karasik (Badoo)Is Apache Kafka already built into your infrastructure? Is it a silver bullet? Unfortunately, not really. Denis Karasik on the example of Badoo will show how to develop Kafka from an experiment in one service to a complete managed-solution, which became the basis for many key tools within the company. The report is primarily for those who are just getting acquainted with Kafka, more sophisticated listeners will be interested in the topic of scaling and managing loaded clusters in this report and the next presentation.
12:00 - Application Design Patterns on Apache Kafka
Alexander Sibiryakov (Scrapinghub)Alexander Sibiryakov will tell you how to design reliable pipelines consisting of a large number of components based on Apache Kafka. The report will analyze architectural patterns of queues, feedback, docking with HTTP, message loss, and processing tasks with a large spread in execution time. After this report, it will become clear how to properly build your pipeline.
13:00 - Monitoring of the modern k8s-project through the eyes of the developer
Sergey Sporyshev (ITSumma)The next keyword in today's technology stack is Kubernetes. It has become the standard for deploying applications on the web, which means it needs to be monitored. Sergey Sporyshev will explain the differences from the usual approaches to monitoring, and will tell you how to approach monitoring organization of a modern project in k8s now. According to insider information, it will affect cool things like monitoring the service mesh level - in general, what is really not enough for most projects.
14:00 - Houston, we have a problem. Failure systems design, Amazon cloud internal service development patterns
Vasily Pantyukhin (Amazon Web Services)Good monitoring warns of problems even before they occur. But it is even more reliable if you initially design the service in such a way as to minimize damage from malfunctions. Design for failure is fundamental in designing Amazon Web Services cloud services. Together with Vasily Pantyukhin, we will deal with this approach, consider some of the reasons for service failures, and talk about what Cell-based architecture, Constant Work, Shuffle Sharding are.
15:00 - Dear DELETE. Typical errors when performing massive operations in highly loaded PostgreSQL databases
Nikolay Samokhvalov (Postgres.ai)Further from tools and development patterns, we move on to working with databases. Nikolay Samokhvalov draws our attention to the consequences of performing massive blocking operations in the database - a highly loaded service will most likely fall. In this report, we learn how Postgres.ai can help and how to configure protection without using it and always act reliably without dropping the prod. All this is based on real experience of pain and huge financial losses.
16:00 - Analysis of queries in MySQL, PostgreSQL, MongoDB
Peter Zaitsev (Percona)Go ahead and think about how to achieve high query performance without DBA. For this task, Peter Zaitsev, one of the nominees for the HighLoad ++ Award, will introduce Percona Monitoring and Management. This is such a tool that helps an ordinary engineer conduct a professional analysis of query performance and more. Nowadays, when a lot of things are left to the hands of frameworks and libraries, as well as in conditions of a shortage in the labor market, the availability and knowledge of such tools is very much in demand.
17:00 - Success and Failure with Redis
Ivan Letenko (Infobip)With PostgreSQL, we’ll switch to technologies that are less common in large production. Infobip has accumulated a lot of interesting things over the 3 years of using Redis. Ivan Letenko will share cases and, most importantly, fails. Let's discuss the organization of distributed caches, queues, the construction of speed limiters and schedulers. Let's talk about transactions on LUA scripts and their performance. We will discuss memory settings, replication, persistence, failover, as well as working with clusters, features of Java drivers and monitoring.
18:00 - HighLoad ++ Prize 2019
It is handed over to professionals who influence the development of the Internet development ecosystem in Russia. This is thanks to the community of professionals for their leaders. For work and dedication, for sleepless nights and countless speeches, for faith in your product or technology.
November 8th
11:00 - Architecture of the Messenger Avito - the path of one message
Alexander Emelin (Avito)11 million unique users per month, 25k RPC requests per second and 500k online connections at the peak - without a doubt, the Avito messenger is a highly loaded project. Alexander Emelin will tell you what services the message goes through before it reaches the addressee. We will find out how the deployment works, how MongoDB and RabbitMQ queues are used. We will discuss the WebSocket protocol and reflect on the need for an HTTP fallback in 2019. We’ll also deal with anti-spam filters and talk about the fault tolerance of this complex system.
12:00 - Personalization in 10 milliseconds, or how tinkoff.ru adapts to you
Vladimir Posvyansky and Ivan Vovk (Tinkoff.ru)Personalization is the scourge of the 21st century, so Tinkoff.ru, to attract users, also decided to personalize pages for individual users using machine learning algorithms. And they say that they can do it in 10 ms - it sounds very promising. Vladimir Posvyansky and Ivan Vovk will introduce us to both the product component and the technical structure of the engine.
13:00 - How to change infrastructure at an explosive growth rate of the company
Nikita Maslyannikov (TsIAN)Closer to dinner with ML, let's jump to DevOps. Nikita Maslyannikov will tell the story of the rapid growth of the TsIAN and related changes in architecture.
Namely, he will dwell in detail on the architecture of monitoring, orchestration, dividing Elasticsearch into microclusters, network setup and SLA calculation. This experience will be useful if you are just working on the architecture or if the current one has begun to cope with the load.
14:00 - Transfer from a TCP needle to UDP with millions of users
Alexander Tobol (Classmates)What to do if you have optimized everything, and the data does not reach the user fast enough? Of course, deal with the network. Alexander Tobol will continue to teach us how to build network protocols and tell how Odnoklassniki transplanted tens of millions of users from TCP to UDP and what it gave. Promised blood, guts and bikes network protocols. Want more hardcore - don't miss this talk.
15:00 - Notified bugs and how to avoid them using the ClickHouse example
Alexey Milovidov (Yandex)If you are writing code, get ready for problems. Without problems in the development of nowhere, they will certainly be expected and it is worth it from all sides - and from your code, and from the compiler, and from the operating system, and from iron. And the larger the scale of the service, the more and more “wonderful” bugs. Alexey Milovidov over the years of development and support for ClickHouse saw different things and on HighLoad ++ will give examples of the most ridiculous, discouraging and hopeless problems. And, of course, he will explain what preventive measures should be taken so that there are fewer problems.
16:00 - 200 TB + Elasticsearch Cluster
Pyotr Zaitsev (Classmates)TCP / UDP is good, but Odnoklassniki also has 4 data centers, 500 machines, 200TB + of data, up to 2 million lines per second at the peak, and 100% service uptime requirements at all costs. Peter Zaitsev will tell how, having all this, build an Elasticsearch cluster for storing logs in a particularly large volume. The task was non-trivial, with a high content of pitfalls. And what happened in the end, we learn at HighLoad ++ 2019.
17:00 - What can a quantum computer do?
Anatoly Dymarsky (Skoltech)At the end of the conference, it seems logical to try to look into the future. Quantum computing promises to revolutionize data processing and change our entire industry. But we need specifics: exactly what tasks can be solved with the help of a quantum computer and when. Anatoly Dymarsky, a theoretical physicist from Skoltech, will answer.
Note that this is only one stream of 16. To freely switch between them and access the video of all reports, you can still buy access to the full broadcast. Or follow the text broadcast in a telegram - it’s already underway, and there we share a lot of interesting insider details and try to convey the atmosphere of the main event of web development in Russia. And in social networks - fb , vk - there will be many, many atmospheric photos. Get connected!