hadoop - Java yarn 应用程序的高可用性

Question

有人可以让我知道确保 yarn 应用高可用性的推荐方法吗？

例如，假设我有3个物理节点。 yarn java应用程序需要3个容器，并且每个物理节点都有足够的资源来容纳3个yarn容器。为确保所有容器都不会落在同一物理节点上而成为“单一故障点”，建议采取什么方法？

Answer 1

这已经在hadoop-mailing list中得到了回答。在此处重新发布答案，以防止链接腐烂

You can use Placements Constraints(https://hadoop.apache.org/docs/r3.1.0/hadoop-yarn/hadoop-yarn-site/PlacementConstraints.html) or set yarn.scheduler.capacity.per-node-heartbeat.maximum-container-assignments to 1 if you RM config with capacityscheduler.

另一种方法-

I think you can use the reservation system in Hadoop 2.8.x as well: https://hadoop.apache.org/docs/r2.8.5/hadoop-yarn/hadoop-yarn-site/ReservationSystem.html