Benjamin Reed wrote:
> It is supported. We run a lot of unit tests that way. Is it possible that the 
> machine is simply getting overloaded. (Especially with respect to disk 
> activity?) You have a rather large timeout, so the machine would really need 
> to be getting swamped.

Hi, Ben.  I'd be very surprised if the machine is getting swamped.  I 
verified this and mpstat shows 98% idle time on both CPUs in the 
machine.  Running the iotop DTrace script shows only two processes 
making disk access within each sampling interval, one of them being 
ZooKeeper.  The disk access time is at most 30ms for ZooKeeper, same for 
the other processes, which computes to 0% (very low) disk utilization 
for each sampling interval.

I will try to debug the single-machine case further if I get a chance. 
I started there because I thought it would be the quickest way to try 
ZooKeeper out.  Turns out deploying it in clustered mode onto 5 virtual 
hosts got me there quicker.  I see no problems in that scenario.

To provide a bit of context, I was working on a locking/leader election 
service for use by applications needing such a service on our research 
cloud computing platform (https://wwww.projectcaroline.net).  Someone in 
the team pointed me to ZooKeeper and rather than reinvent the wheel, I 
am trying to see if we can use ZooKeeper "out-of-the-box".  So far it 
seems to fit the bill.  Given none of the deployment scenarios we have 
involve client and service on the same host (in fact, such a deployment 
is not possible in our platform), the exceptions I see when running 
client and service on the same machine are not an impediment.  I am 
curious about them, so I will probably investigate further as time permits.

Thanks again for your help.

Regards,

Juan


> 
> ben
> 
> On Friday 23 May 2008 13:00:02 Juan Ramirez wrote:
>> A piece of info I left out which appears to be relevant is that I was
>> running the standalone server and its client on the same machine.
>> Perhaps this is not supported?  I am not seeing the problem in a
>> clustered deployment with 5 zookeeper instances.
>>
>> Thanks,
>>
>> Juan
>>
>> Juan Ramirez wrote:
>>> Mahadev Konar wrote:
>>>> Hi Juan,
>>>>   Can you tell us what you are doing in your client code and when you
>>>> see this problem?
>>> Hi, Mahadev.  It seems to happen when attempting to create a new node.
>>> Might be a timing issue because it does not happen all the time.  The
>>> client code where this happens is quite simple:
>>>
>>>      if (keeper.exists(path, false) == null) {
>>>          file = keeper.create(path, "hello".getBytes(),
>>>              ZooDefs.Ids.OPEN_ACL_UNSAFE, flags);
>>>
>>>  >You are running with a standalone zookeeper server
>>>> right?
>>> Correct.
>>>
>>> Thanks,
>>>
>>> Juan
>>>
>>>
>>> -------------------------------------------------------------------------
>>> This SF.net email is sponsored by: Microsoft
>>> Defy all challenges. Microsoft(R) Visual Studio 2008.
>>> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
>>> _______________________________________________
>>> Zookeeper-user mailing list
>>> Zookeeper-user@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/zookeeper-user
>> -------------------------------------------------------------------------
>> This SF.net email is sponsored by: Microsoft
>> Defy all challenges. Microsoft(R) Visual Studio 2008.
>> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
>> _______________________________________________
>> Zookeeper-user mailing list
>> Zookeeper-user@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/zookeeper-user
> 
> 


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Zookeeper-user mailing list
Zookeeper-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/zookeeper-user

Reply via email to