FLINK WEEKLY 2019/40

[tison]

FLINK WEEKLY 2019/40 <https://zhuanlan.zhihu.com/p/85542386>用户问题

Dynamic stream handling
<https://lists.apache.org/x/thread.html/cefcd93d889b3d29c0d9d0c01a91be77d359b470d4af29eb4f1f9807@%3Cuser-zh.flink.apache.org%3E>

FLINK 暂不支持流图的动态更新，但这是 FLINK 计划中支持的功能

[SURVEY] What is the most subtle/hard to catch bug that people have seen?
<https://lists.apache.org/x/thread.html/8a85b99c214a2e2185dab4da105540cb35dbb5e0cc4e958b8ae241c1@%3Cuser.flink.apache.org%3E>

Konstantinos Kallas 发起了一个有趣的调查，关于 FLINK 用户遇到过的最微妙棘手的问题。他和他的团队正准备搭建一个 FLINK
的测试框架，希望能够收集已有的问题的样本

Broadcast state
<https://lists.apache.org/x/thread.html/4afb03bee2ac1c6ca004f0eb5f1efbacb661137e45c25e06818c5303@%3Cuser.flink.apache.org%3E>

关于在作业中 Broadcast state 的复用问题

Finding the Maximum Value Received so far in a Stream
<https://lists.apache.org/x/thread.html/dca0212b1f473c0b72720cccf1b6a227e1654e06449664d702b2ec34@%3Cuser.flink.apache.org%3E>

场景实现，查找流中当前的最大值

POJO serialization vs immutability
<https://lists.apache.org/x/thread.html/66cb883c6ad1a0d59f0056e9ab77b160aaf9d2c95baaeb28008f8c7a@%3Cuser.flink.apache.org%3E>

关于 FLINK 中 POJO 实现的细节，由于 POJO 的域是可变的，所以在默认的 hashCode 实现下不能用作键值对的键
已知缺陷

FLINK-14315 NPE with JobMaster.disconnectTaskManager
<https://issues.apache.org/jira/browse/FLINK-14315>

JobMaster 的竞态条件使得运行中可能抛出空指针异常，已定位到问题，将在 1.10.0/1.9.1/1.8.3 中修复
开发讨论

[SURVEY] Dropping non Credit-based Flow Control
<https://lists.apache.org/x/thread.html/4ffa21acd88cb909fc09fc052ef290e323bc057158a60e52aa8104f1@%3Cdev.flink.apache.org%3E>

Piotr Nowojski 发起了废除非 Credit-based 的流量控制机制的讨论。在 FLINK 1.5 中引入了 Credit-based
的流量控制机制，目前 FLINK 的网络栈正在活跃发展，废除这部分代码将有利于开发的进行

[jira] [Created] (FLINK-14320) [FLIP-66] Support Time Attribute in SQL DDL
<https://lists.apache.org/x/thread.html/fd93b20688cc1db821f0ebc3be1406a2c360676f3aa6c67bb8318473@%3Cdev.flink.apache.org%3E>

Jark Wu 的 FLIP-66 已经通过投票，开始开发。FLIP-66 旨在提供 SQL DDL 中对时间属性的支持

[DISCUSS] Improve Flink logging with contextual information
<https://lists.apache.org/x/thread.html/91940af79bc25440ac269af30cd46ea28fed187b31fa142a5c457739@%3Cdev.flink.apache.org%3E>

Gyula Fóra 发起了关于丰富 FLINK 日志内容的讨论，主要是提供关于 TaskManager/Container/JobId 等信息

[DISCUSS] FLIP-65: New type inference for Table API UDFs
<https://lists.apache.org/x/thread.html/2b8cfd811a927d64a79f47387b7412c3b98a11fcb9358d3c23ef666c@%3Cdev.flink.apache.org%3E>

Timo Walther 的 FLIP-65 旨在为 Table API 的用户定义函数提供新的类型接口，这也是新一轮 Table API
开发中的一部分

[DISCUSS] FLIP-76: Unaligned checkpoints
<https://lists.apache.org/x/thread.html/a2b58b7b2b24b9bd4814b2aa51d2fc44b08a919eddbb5b1256be5b6a@%3Cdev.flink.apache.org%3E>

Arvid Heise 的 FLIP-76 引入了 Unaligned checkpoints，旨在优化背压情况下的 checkpoint 性能

[VOTE] FLIP-73: Introducing Executors for job submission
<https://lists.apache.org/x/thread.html/0a76dc56d4201d65dd1d6e47b22306a37acf1d0ec6e2db8c144e3541@%3Cdev.flink.apache.org%3E>
[VOTE]
FLIP-74: Flink JobClient API
<https://lists.apache.org/x/thread.html/a7096ff22025ea5bb87fb7d3d36f2fe07f663472f4ed4fdad012d2e1@%3Cdev.flink.apache.org%3E>

Client API 改进的两个 FLIP 进入投票阶段
社区发展

[VOTE] Release 1.9.1, release candidate #1
<http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/VOTE-Release-1-9-1-release-candidate-1-td33637.html>

Jark Wu 作为 1.9.1 的 release manager 拉出了第一个 release candidate

Real-time experiment analytics at Pinterest using Apache Flink
<https://medium.com/pinterest-engineering/real-time-experiment-analytics-at-pinterest-using-apache-flink-841c8df98dc2>

来自 Pinterest Engineering 的开发者分享了他们使用 FLINK 做实时计算的经验

Turning messy data into a gold mine using Spark, Flink, and ScyllaDB
<https://www.dynamicyield.com/blog/turning-messy-data-into-a-gold-mine/>

来自 DynamicYield 的 Oran Hirsch 分享了他们基于 Spark Flink ScyllaDB 的数据分析栈