You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@curator.apache.org by Adarsh Bhat <ad...@gmail.com> on 2013/09/17 21:09:19 UTC
Possible bug in DistributedIdQueue recipe
Hi curator users!
I'm using the DistributedIdQueue recipe in my system. Although I'm
aware of the tech note that this is not suitable for production use,
this is a pretty lightweight use case with very few items in the queue
at any time. It has actually been working rather well.
I encountered a situation recently that was uncovered because of a bug
in my code. I neglected to catch an exception in my QueueConsumer,
leading to an endless re-queueing of the item. While this is perhaps
expected behavior, I also noticed that the name of the znode (and the
corresponding lock znode) grew with each requeue. After many hours,
this name grew so long that it brought the entire zookeeper ensemble
to its knees. It refused to accept any client connections.
Starting with this:
Exception processing queue item: queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f
And this is how the node name grew:
queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|00000010940000001095
queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|0000001094000000109500000010960000001097
queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|000000109400000010950000001096000000109700000010980000001099
...and so on
Before I knew it the node name was many pages long.
Is this a defect in the recipe? Or have I done anything wrong myself?
Regards
Adarsh Bhat
Re: Possible bug in DistributedIdQueue recipe
Posted by Adarsh Bhat <ad...@gmail.com>.
DistributedQueue znode name grows on re-queue
https://issues.apache.org/jira/browse/CURATOR-56
I have created a jira entry and attached a patch that includes a
failing test case. Please note that the bug is really in the
DistributedQueue recipe. I just happened to be using a
DistributedIdQueue.
Thanks!
Adarsh Bhat
HP Software
On Tue, Sep 17, 2013 at 12:13 PM, Jordan Zimmerman
<jo...@jordanzimmerman.com> wrote:
> That can't be correct behavior. Please open a Jira with a test case.
>
> ====================
> Jordan Zimmerman
>
> On Sep 17, 2013, at 2:09 PM, Adarsh Bhat <ad...@gmail.com> wrote:
>
>> Hi curator users!
>>
>> I'm using the DistributedIdQueue recipe in my system. Although I'm
>> aware of the tech note that this is not suitable for production use,
>> this is a pretty lightweight use case with very few items in the queue
>> at any time. It has actually been working rather well.
>>
>> I encountered a situation recently that was uncovered because of a bug
>> in my code. I neglected to catch an exception in my QueueConsumer,
>> leading to an endless re-queueing of the item. While this is perhaps
>> expected behavior, I also noticed that the name of the znode (and the
>> corresponding lock znode) grew with each requeue. After many hours,
>> this name grew so long that it brought the entire zookeeper ensemble
>> to its knees. It refused to accept any client connections.
>>
>> Starting with this:
>> Exception processing queue item: queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f
>>
>> And this is how the node name grew:
>> queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|00000010940000001095
>> queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|0000001094000000109500000010960000001097
>> queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|000000109400000010950000001096000000109700000010980000001099
>> ...and so on
>>
>> Before I knew it the node name was many pages long.
>>
>> Is this a defect in the recipe? Or have I done anything wrong myself?
>>
>> Regards
>> Adarsh Bhat
Re: Possible bug in DistributedIdQueue recipe
Posted by Jordan Zimmerman <jo...@jordanzimmerman.com>.
That can't be correct behavior. Please open a Jira with a test case.
====================
Jordan Zimmerman
On Sep 17, 2013, at 2:09 PM, Adarsh Bhat <ad...@gmail.com> wrote:
> Hi curator users!
>
> I'm using the DistributedIdQueue recipe in my system. Although I'm
> aware of the tech note that this is not suitable for production use,
> this is a pretty lightweight use case with very few items in the queue
> at any time. It has actually been working rather well.
>
> I encountered a situation recently that was uncovered because of a bug
> in my code. I neglected to catch an exception in my QueueConsumer,
> leading to an endless re-queueing of the item. While this is perhaps
> expected behavior, I also noticed that the name of the znode (and the
> corresponding lock znode) grew with each requeue. After many hours,
> this name grew so long that it brought the entire zookeeper ensemble
> to its knees. It refused to accept any client connections.
>
> Starting with this:
> Exception processing queue item: queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f
>
> And this is how the node name grew:
> queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|00000010940000001095
> queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|0000001094000000109500000010960000001097
> queue-|616682fa-15b6-48e0-b7cc-ea6e3023248f|000000109400000010950000001096000000109700000010980000001099
> ...and so on
>
> Before I knew it the node name was many pages long.
>
> Is this a defect in the recipe? Or have I done anything wrong myself?
>
> Regards
> Adarsh Bhat