For 2.0.0 at least, for controlling DN heartbeat interval timeouts
(missing which gets it marked dead at NN), tweak the prop:
"dfs.namenode.heartbeat.recheck-interval" (default 5 minutes in ms).
The actual timeout to measure is determined from this value, but isn't
exactly this. But lowering it will get you what you need.
On Mon, Jun 11, 2012 at 7:02 PM, Giulio D'Ippolito
<[EMAIL PROTECTED]> wrote:
> Thanks for the answer,
> The demon of mapreduce is not running (i guess that the mapreduce config is
> not readed as well), it is just the dfs running, more specifically 1
> namenode and 3 datanodes. I would like that after a prefixed time of
> disconnection the node is considered lost. Putting the line you cite in the
> hdfs-core.xml does not change as well.
> 2012/6/11 madhu phatak <[EMAIL PROTECTED]>
>> mapred.task.timeout is set to 10 mins which determines how much time have
>> to waited before a task is considered as timed out.
>> On Mon, Jun 11, 2012 at 6:28 PM, Giulio D'Ippolito
>> <[EMAIL PROTECTED]> wrote:
>>> Hello, I'm running hadoop-0.22.0 with the RAID extension, I'm having
>>> trouble to find the parameter that control the time before a node is
>>> considered as disconnected. The default value is set to 10 minutes, I would
>>> like to decrease a bit. I didn't find a way googling.