[ https://issues.apache.org/jira/browse/NIFI-2364?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15403180#comment-15403180 ]
Koji Kawamura edited comment on NIFI-2364 at 8/2/16 1:35 AM: ------------------------------------------------------------- This issue happened after applying NIFI-2363 and testing with a clustered NIFI. Without NIFI-2363, all nodes will respond with the same response, then this will not happen. So, this issue is more a sub task of NIFI-2363. Based on Joe's review comment, I thought we will need more time to go though review & test cycle. Since this is a new feature, I'm ok with moving it to 1.1. Also, cancel the patch to break it down to smaller PRs only focusing individual JIRA issue. was (Author: ijokarumawak): This issue happened after applying NIFI-2363 and testing with a clustered NIFI. Without NIFI-2363, all nodes will respond with the same response, and this will not happen. So, this issue is more a sub task of NIFI-2363. Spoked with Joe Witt and Joe Percivall about release target. Since there is not enough time to go though review & test cycle to be in 1.0. Also, cancel the patch to break it down to smaller PRs only focusing individual JIRA issue. > Node shouldn't be disconnected from a cluster when clear component state > operation gets error related to underlying data source > ------------------------------------------------------------------------------------------------------------------------------- > > Key: NIFI-2364 > URL: https://issues.apache.org/jira/browse/NIFI-2364 > Project: Apache NiFi > Issue Type: Sub-task > Components: Core Framework > Reporter: Koji Kawamura > Assignee: Koji Kawamura > Fix For: 1.1.0 > > > NodeClusterCoordinator.afterRequest() disconnects a node if it couldn't > handle mutation requests such as POST, assuming that node is not functioning > properly. > However, if the cause of issue is related to an external data source, such as > Zookeeper or Kafka, other node will have the same problem even if other node > is elected as a primary node. In such case, disconnecting node wouldn't be a > good recovery solution. Instead, the node should keep connected, and let > users retry the operation. -- This message was sent by Atlassian JIRA (v6.3.4#6332)