Thanks.   So that is the reason why Sharninder thought it might be the JVM ?

What do you suggest to look for in this case ?

thanks,
-Gary


On Sep 20, 2014, at 8:47 PM, Hari Shreedharan <[email protected]> wrote:

> The shutdown hook is triggered when the app is shutdown. Since Flume does not 
> shutdown by itself, some other external process must have sent it a SIGTERM 
> (in a SIGKILL case, the shutdown hook won't execute).
> 
> On Sat, Sep 20, 2014 at 7:20 PM, terreyshih <[email protected]> wrote:
> Yes, I have increased to 1G.  I have also started the jmx remote monitoring 
> on the jvm which I don’t see any issue.
> 
> Do you know when the shut down hook kicks in ?  (under what conditions ?)
> 
> thx
> 
> On Sep 20, 2014, at 7:01 PM, Sharninder <[email protected]> wrote:
> 
>> Have you increased the JVM heap size or running with the default 20M only?
>> 
>> --
>> Sharninder
>> 
>> On 21-Sep-2014, at 3:08 am, terreyshih <[email protected]> wrote:
>> 
>>> BTW, no exceptions are thrown.  It just started the agent-shutdown-hook.
>>> 
>>> thx
>>> 
>>> On Sep 20, 2014, at 2:33 PM, terreyshih <[email protected]> wrote:
>>> 
>>>> Hi,
>>>> 
>>>> I am having a lot of difficulty trying to debug an issue.  The flume 
>>>> server stops working after a few hours and the JVM shuts down.  The 
>>>> snippet of the message I got in the log starts with following:
>>>> 
>>>> 
>>>> 20 Sep 2014 12:24:10,899 DEBUG [New I/O  worker #9] 
>>>> (org.hbase.async.RegionClient.decode:1349)  - ------------------<< LEAVING 
>>>>  DECODE <<------------------ time elapsed: 75us
>>>> 20 Sep 2014 12:24:19,942 INFO  [agent-shutdown-hook] 
>>>> (org.apache.flume.lifecycle.LifecycleSupervisor.stop:79)  - Stopping 
>>>> lifecycle supervisor 13
>>>> 20 Sep 2014 12:24:19,943 INFO  [agent-shutdown-hook] 
>>>> (org.apache.flume.instrumentation.MonitoredCounterGroup.stop:149)  - 
>>>> Component type: CHANNEL, name: myChannelC stopped
>>>> 
>>>> 
>>>> From there it starts to shut down all the channels and eventually the JVM 
>>>> shuts down.
>>>> 
>>>> I am running the agent in debug mode and I can see my data coming in 
>>>> correctly for a couple of hours and the snippet shows, flume enters the 
>>>> “LEAVING DECODE” section of the flume debug engine.  I am writing the data 
>>>> to HBase sinks by using AsycnHBase.
>>>> 
>>>> I would appreciate it if someone can help me with this.  This is 
>>>> reproducible around every 2 hours.
>>>> 
>>>> When does the agent-shutdown-hook gets triggered ?
>>>> 
>>>> thanks
>>> 
> 
> 

Reply via email to