I don't think that your explanation around "max.in.flight.requests.per.conection" is good enough. It's a far more complicated setting with regards to order. Otherwise the presentation was fabulous! It took a while to understand your point about : record-queue-time-avg/2. If you have 4 messages and in a perfect world they would arrive 2ms apart (for example), it means that the batch will have stayed 6ms in the record accumulator (again, in a perfect world). At the same time the average of each message in a batch would be : (6ms for the first message) + (4ms for the second) + (2ms for the third) + (0ms for the fourth) = 12ms / 4 messages = 3ms; OR record-queue-time-avg/2.
Лучший доклад по Кафке, что я смотрел. Вот бы ещё по брокерам аналогичный разбор!
Очень классный доклад, спасибо!
Было-бы посмотреть похожий разбор по реактивному клиенту.
Реактивный клиент это же просто обертка над обычным, разве нет?
@@virus0936"обёртка" - да, "просто" - нет.
I don't think that your explanation around "max.in.flight.requests.per.conection" is good enough. It's a far more complicated setting with regards to order. Otherwise the presentation was fabulous!
It took a while to understand your point about : record-queue-time-avg/2. If you have 4 messages and in a perfect world they would arrive 2ms apart (for example), it means that the batch will have stayed 6ms in the record accumulator (again, in a perfect world). At the same time the average of each message in a batch would be : (6ms for the first message) + (4ms for the second) + (2ms for the third) + (0ms for the fourth) = 12ms / 4 messages = 3ms; OR record-queue-time-avg/2.