[ https://issues.apache.org/jira/browse/HDFS-7858?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14341375#comment-14341375 ]
Arun Suresh commented on HDFS-7858: ----------------------------------- [~kihwal], [~bikassaha], I understand your concerns over the use of ZK. But consider the following: # Most DFSClients are cached (since the Filesystem objects are generally cached). Thus long lived clients will probably not have more than 1-2 persistent connections to ZK. # Short lived clients will first check if there is cached entry (possibly in the home directory, something like ~/.lastNN) that contains the last accessed active NN. The client will proceed to connect to that NN first (thereby removing non-determinism from the current scheme).. and will most probably succeed. It will contact ZK only if the connection was unsuccessful.. and we can limit this to just a ping (not a watch registration) so the connection is not persistent. # A Client that has connected to a NN without the need for a ZK connection can continue to NOT talk to ZK till ## the client dies and no ZK connection is ever made ## after waiting for a configurable time (maybe an hour).. after which it is established that it is a ling lived client, Do you think this might reduce the total number of connection to ZK at any point of time ? > Improve HA Namenode Failover detection on the client using Zookeeper > -------------------------------------------------------------------- > > Key: HDFS-7858 > URL: https://issues.apache.org/jira/browse/HDFS-7858 > Project: Hadoop HDFS > Issue Type: Improvement > Reporter: Arun Suresh > Assignee: Arun Suresh > > In an HA deployment, Clients are configured with the hostnames of both the > Active and Standby Namenodes.Clients will first try one of the NNs > (non-deterministically) and if its a standby NN, then it will respond to the > client to retry the request on the other Namenode. > If the client happens to talks to the Standby first, and the standby is > undergoing some GC / is busy, then those clients might not get a response > soon enough to try the other NN. > Proposed Approach to solve this : > 1) Since Zookeeper is already used as the failover controller, the clients > could talk to ZK and find out which is the active namenode before contacting > it. > 2) Long-lived DFSClients would have a ZK watch configured which fires when > there is a failover so they do not have to query ZK everytime to find out the > active NN > 2) Clients can also cache the last active NN in the user's home directory > (~/.lastNN) so that short-lived clients can try that Namenode first before > querying ZK -- This message was sent by Atlassian JIRA (v6.3.4#6332)