[jira] [Assigned] (HDFS-13633) RBF: Synchronous way to create RPC client connections to NN
[ https://issues.apache.org/jira/browse/HDFS-13633?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] CR Hota reassigned HDFS-13633: -- Assignee: CR Hota (was: CR Hota(invalid)) > RBF: Synchronous way to create RPC client connections to NN > --- > > Key: HDFS-13633 > URL: https://issues.apache.org/jira/browse/HDFS-13633 > Project: Hadoop HDFS > Issue Type: Sub-task > Components: federation >Reporter: CR Hota >Assignee: CR Hota >Priority: Major > > Currently the router code does the following. > # IPC handler thread gets a connection from the pool, even if the connection > is NOT usable. > # At the same time the IPC thread also submits a request to connection > creator thread for adding a new connection to the pool asynchronously. > # The new connection is NOT utilized by the IPC threads that get back an > unusable connection. > With this approach burst behaviors of clients, fill up the pool without > necessarily using the connections. Also the approach is indeterministic. > We propose a flag that can allow router admins to control the behavior of > getting connections by the IPC handler threads. The flag would allow to > toggle ON/OFF asynchronous vs synchronous way of connection creation. > In the new model, if a connection is unusable, IPC handler thread would go > ahead and create a connection and add to the pool and utilize it > subsequently. It would still utilize the unusable connection if the pool is > full. > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Assigned] (HDFS-13633) RBF: Synchronous way to create RPC client connections to NN
[ https://issues.apache.org/jira/browse/HDFS-13633?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Íñigo Goiri reassigned HDFS-13633: -- Assignee: CR Hota > RBF: Synchronous way to create RPC client connections to NN > --- > > Key: HDFS-13633 > URL: https://issues.apache.org/jira/browse/HDFS-13633 > Project: Hadoop HDFS > Issue Type: Sub-task > Components: federation >Reporter: CR Hota >Assignee: CR Hota >Priority: Major > > Currently the router code does the following. > # IPC handler thread gets a connection from the pool, even if the connection > is NOT usable. > # At the same time the IPC thread also submits a request to connection > creator thread for adding a new connection to the pool asynchronously. > # The new connection is NOT utilized by the IPC threads that get back an > unusable connection. > With this approach burst behaviors of clients, fill up the pool without > necessarily using the connections. Also the approach is indeterministic. > We propose a flag that can allow router admins to control the behavior of > getting connections by the IPC handler threads. The flag would allow to > toggle ON/OFF asynchronous vs synchronous way of connection creation. > In the new model, if a connection is unusable, IPC handler thread would go > ahead and create a connection and add to the pool and utilize it > subsequently. It would still utilize the unusable connection if the pool is > full. > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
