Le 2010-05-25 à 13:28, Chuck Hill a écrit :

> Hi Pascal,
> 
> On May 24, 2010, at 5:14 PM, Pascal Robert wrote:
> 
>> Hi everyone,
>> 
>> I'm working on a Nagios plugin that will use the /admin/info direct action 
>> that was added in the Wonder variant of JavaMonitor. Right now, the plugin 
>> is doing the doing the following :
>> 
>>      - for the "state" key, if the state is DEAD, CRASHING or STOPPING, it 
>> sends a CRITICAL signal, if it's UNKNOWN or STARTING, it will send a WARNING 
>> signal, if the state is ALIVE, it will be OK.
> 
> I think that STOPPING would indicate a manual shutdown or a scheduled 
> restart.  Do you really want to send a notification in that case?

I think a notification should be send, unless you're doing something big at 
shutdown, you might never get a notification. For example, our Nagios setup is 
sending notifications if the service is having a problem after 5 minutes 
(regular interval is 3 minutes, interval when a problem is detected is 1 
minute, so 3 + 1 + 1). If your app is in a STOPPING state for more than 4 
minutes, it might be a problem. But a warning signal should be better than a 
critical one.

> UNKNOWN probably represents a timeout talking to wotaskd which would indicate 
> one:
> - deadlocked / backlogged instance
> - wotaskd stopped on some machine
> - network problems reaching some machine
> 
> Those might warrant a CRITICAL signal.

Good catch.

> 
>>      - for the "deaths", "transactions", "activeSessions", 
>> "averageIdlePeriod" and "avgTransactionTime" keys, it will check against the 
>> warning and critical values, if the actual values are higher than the 
>> params, it will send the appropriate signals.
> 
> The last time I looked, these are not cleared when an application is 
> unscheduled and stopped.  If these values were outside of the limits when 
> this happened, that could result in a lot of notifications.
> 
> 
>>      - for refusingNewSessions, scheduled and autoRecover, it will send a 
>> WARNING signal if the response is "false"
> 
> I often (usually) have configured, but not scheduled instances for use when 
> upgrading to a new version, handling server failover, higher loads etc.  Is 
> there a way to define only the instances that are expected to be scheduled?

Frankly, I don't know how to handle this. Right now, the plugin works on a 
specific instance. I guess we can pass an array of instance IDs, and if the 
specified key to check is reaching a warning or critical level for all 
instances, it will send a notification.

> Warning on refusingNewSessions is going to send notifications for scheduled 
> restarts, probably not what you want.

Will add a condition for this, if scheduled is true, refusingNewSessions will 
be ok if set to false.

> 
>> Any opinions on this? The only thing that I need to work on is the help 
>> output, so if you want to try the plugin (it's a PERL script), send me a 
>> note. You need any version of Nagios and the Wonder variant of JavaMonitor.
> 
> 
> Sounds like it could be  useful.
> 
> One other thing is that sending passwords on the URL is insecure and 
> passwords that contain non-URL friendly characters are a problem (they don't 
> seem to get decoded in JavaMonitor, not sure about that, I just changed the 
> password).

Will try that. _______________________________________________
Do not post admin requests to the list. They will be ignored.
Webobjects-dev mailing list      ([email protected])
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/webobjects-dev/archive%40mail-archive.com

This email sent to [email protected]

Reply via email to