12:42 p.m.
When a worker is killed because the virtaulized instance on it is shut
down, the mod_cluster_manager app still displays the worker as healthy
for about 5 minutes before removing it. However, I confirmed that the
worker is dead by pinging it (which failed).
Is there a timeout parameter that I can set to have the httpd proxy
remove a failed worker sooner ? KeepAliveTimeout apparently is used for
closing idle connections, but not for detecting dead workers, so this is
not the option I'm looking for.
Besides: isn't the JBoss master supposed to unregister failed workers ?
--
Bela Ban
Lead JGroups / Clustering Team
JBoss
12:57 p.m.
On 06/08/2010 12:42 PM, Bela Ban wrote:
When a worker is killed because the virtaulized instance on it is
shut
down, the mod_cluster_manager app still displays the worker as healthy
for about 5 minutes before removing it. However, I confirmed that the
worker is dead by pinging it (which failed).
The worker should be marked with Status: NOK after about 30 seconds but
only removed after 5 minutes.
Is there a timeout parameter that I can set to have the httpd proxy
remove a failed worker sooner ?
There isn't yet a directive for that: httpd tries to cping the node 60
time and remove it after the last try.
Cheers
Jean-Frederic
1:08 p.m.
jean-frederic clere wrote:
On 06/08/2010 12:42 PM, Bela Ban wrote:
> When a worker is killed because the virtaulized instance on it is shut
> down, the mod_cluster_manager app still displays the worker as healthy
> for about 5 minutes before removing it. However, I confirmed that the
> worker is dead by pinging it (which failed).
The worker should be marked with Status: NOK after about 30 seconds
Can I configure the 30 secs, or is this hard coded ?
but only removed after 5 minutes.
OK, got it. This means though that a worker marked as NOTOK won't get
any requests, correct ?
> Is there a timeout parameter that I can set to have the httpd proxy
> remove a failed worker sooner ?
There isn't yet a directive for that: httpd tries to cping the node 60
time and remove it after the last try.
OK. Is this (the 60 times) configurable ?
--
Bela Ban
Lead JGroups / Clustering Team
JBoss
1:22 p.m.
On 06/08/2010 01:08 PM, Bela Ban wrote:
jean-frederic clere wrote:
> On 06/08/2010 12:42 PM, Bela Ban wrote:
>> When a worker is killed because the virtaulized instance on it is shut
>> down, the mod_cluster_manager app still displays the worker as healthy
>> for about 5 minutes before removing it. However, I confirmed that the
>> worker is dead by pinging it (which failed).
>
> The worker should be marked with Status: NOK after about 30 seconds
Can I configure the 30 secs, or is this hard coded ?
hmm after checking it should be 5 seconds. That is LBstatusRecalTime.
(See http://www.jboss.org/mod_cluster/native/config.html)
> but only removed after 5 minutes.
OK, got it. This means though that a worker marked as NOTOK won't get
any requests, correct ?
Yes.
>
>> Is there a timeout parameter that I can set to have the httpd proxy
>> remove a failed worker sooner ?
>
> There isn't yet a directive for that: httpd tries to cping the node 60
> time and remove it after the last try.
OK. Is this (the 60 times) configurable ?
No.
Cheers
Jean-Frederic
5322
days inactive
5322
days old
mod_cluster-dev@lists.jboss.org
3 comments
2 participants
participants (2)
-
Bela Ban -
jean-frederic clere