[ https://issues.apache.org/jira/browse/ZOOKEEPER-417?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12724309#action_12724309 ]
Benjamin Reed commented on ZOOKEEPER-417: ----------------------------------------- the release audit generates warnings before the way i added the new keeper exception codes. we made the integers deprecated, so i've made the new ones deprecated as well. should i put the integer in the new error code rather than added a new deprecated constant? i can't find any failed tests in the test results. what am i missing? > stray message problem when changing servers > ------------------------------------------- > > Key: ZOOKEEPER-417 > URL: https://issues.apache.org/jira/browse/ZOOKEEPER-417 > Project: Zookeeper > Issue Type: Bug > Reporter: Benjamin Reed > Assignee: Benjamin Reed > Priority: Blocker > Fix For: 3.2.0 > > Attachments: ZOOKEEPER-417.patch, ZOOKEEPER-417.patch > > > There is a possibility for stray messages from a previous connection to > violate ordering and generally cause problems. Here is a scenario: we have a > client, C, two followers, F1 and F2, and a leader, L. The client is connected > to F1, which is a slow follower. C sends setData("/a", "1") to F1 and then > loses the connection, so C reconnects to F2 and sends setData("/a", "2"). it > is possible, if F1 is slow enough and the setData("/a", "1") got onto the > network before the connection break, for F1 to forward the setData("/a", "1") > to L after F2 forwards setData("/a", "2"). > to fix this, the leader should keep track of which follower last registered a > session for a client and drop any requests from followers for clients for > whom they do not have a registration. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.