[
https://issues.apache.org/jira/browse/HDFS-4911?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Colin Patrick McCabe updated HDFS-4911:
---------------------------------------
Attachment: HDFS-4911.001.patch
* reduce DFS_CLIENT_SOCKET_CACHE_EXPIRY_MSEC_DEFAULT to 3 seconds
* bump up DFS_DATANODE_SOCKET_REUSE_KEEPALIVE_DEFAULT to 4 seconds
* PeerCache now checks for expired entries when you try to get an entry out of
it, and will retry if so.
* TestDataTransferKeepalive: rename testKeepaliveTimeouts to
testDatanodeRespectsKeepAliveTimeout and create
testClientResponsesKeepAliveTimeout. De-globalize dfsClient to allow us to use
clients with different settings.
> Reduce PeerCache timeout to be commensurate with
> dfs.datanode.socket.reuse.keepalive
> ------------------------------------------------------------------------------------
>
> Key: HDFS-4911
> URL: https://issues.apache.org/jira/browse/HDFS-4911
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Colin Patrick McCabe
> Assignee: Colin Patrick McCabe
> Priority: Minor
> Attachments: HDFS-4911.001.patch
>
>
> The default timeout for the client's PeerCache is much longer than the
> DataNode would possibly keep the socket open. Specifically,
> {{dfs.client.socketcache.expiryMsec}} defaults to 2 *
> 60 * 1000 (2 minutes), whereas {{dfs.datanode.socket.reuse.keepalive}}
> defaults to 1000 (1 second). We should make these more similar to minimize
> situations where the client tries to use sockets which have gone stale.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
