You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-dev@hadoop.apache.org by "dhruba borthakur (JIRA)" <ji...@apache.org> on 2009/06/23 11:14:07 UTC
[jira] Created: (HADOOP-6099) Allow configuring the IPC module to
send pings
Allow configuring the IPC module to send pings
----------------------------------------------
Key: HADOOP-6099
URL: https://issues.apache.org/jira/browse/HADOOP-6099
Project: Hadoop Common
Issue Type: Improvement
Components: ipc
Reporter: dhruba borthakur
Assignee: dhruba borthakur
The IPC Client sets a socketTimeout for the time period specified by the pingInterval and then sends a ping every pingInterval. This means that if a RPC server does not respond to a RPC client, then the RPC client blocks forever. This is a problem for applications that wants to switch quickly from one un-responsive HDFS cluster to a good one.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HADOOP-6099) Allow configuring the IPC module to
send pings
Posted by "dhruba borthakur (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HADOOP-6099?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12725099#action_12725099 ]
dhruba borthakur commented on HADOOP-6099:
------------------------------------------
Konstantin commented on this issue via HDFS-278:
"why do you need additional configuration parameter ipc.client.ping?
Can you just use set ipc.ping.interval = -1 and treat it as no ping?
Then by default since ipc.ping.interval parameter is unofficial it will still equal 1 min,
and if you want to turn pings off you will need to specify it. Everything looks consistent with your plan.
I mean less parameters is better, right?"
I agree that your suggestion is simpler. But the reason I need another parameter is because I would like the dfsclient to abort if the server is unresponsive for only 20 seconds or so. Your approach does not let me change the default timeout value of 1 minute. Thats the reason I need a separate config parameter. Does it make sense?
> Allow configuring the IPC module to send pings
> ----------------------------------------------
>
> Key: HADOOP-6099
> URL: https://issues.apache.org/jira/browse/HADOOP-6099
> Project: Hadoop Common
> Issue Type: Improvement
> Components: ipc
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
> Attachments: ipcPing.txt
>
>
> The IPC Client sets a socketTimeout for the time period specified by the pingInterval and then sends a ping every pingInterval. This means that if a RPC server does not respond to a RPC client, then the RPC client blocks forever. This is a problem for applications that wants to switch quickly from one un-responsive HDFS cluster to a good one.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HADOOP-6099) Allow configuring the IPC module to
send pings
Posted by "dhruba borthakur (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HADOOP-6099?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
dhruba borthakur updated HADOOP-6099:
-------------------------------------
Attachment: ipcPing.txt
If ipc.client.ping is set to false, then the ipc Client does not send periodic pings to IPC server. The default is set to true to maintain compatibility with pre-existing semantics of sending periodic pings.
> Allow configuring the IPC module to send pings
> ----------------------------------------------
>
> Key: HADOOP-6099
> URL: https://issues.apache.org/jira/browse/HADOOP-6099
> Project: Hadoop Common
> Issue Type: Improvement
> Components: ipc
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
> Attachments: ipcPing.txt
>
>
> The IPC Client sets a socketTimeout for the time period specified by the pingInterval and then sends a ping every pingInterval. This means that if a RPC server does not respond to a RPC client, then the RPC client blocks forever. This is a problem for applications that wants to switch quickly from one un-responsive HDFS cluster to a good one.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.