You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@zookeeper.apache.org by Vitalii Tymchyshyn <ti...@gmail.com> on 2011/11/17 10:59:26 UTC
Get sequence number without node creation
Hello.
I have a feature request I'd like to discuss: How about a call that will
give next sequential node number without node creation. Main reason is
that in many cases this would allow retries without leak on reconnect, e.g.:
Now:
1) create(SEQUENTIAL) - disconnect - retry - and you've just leaked a node
Can be
2) incrementCounter() - create(name). And you can make a retry in any
place. The worst thing that can occur is that you will have a hole in
your numbering
Another cases this can help are simple counters (no need to create nodes
at all) or simply different sequential nodes naming schema.
Best regards, Vitalii Tymchyshyn
Re: Get sequence number without node creation
Posted by nileader <ni...@gmail.com>.
No。any time，two return result will be by two update request.
2011/11/18 Vitalii Tymchyshyn <ti...@gmail.com>
> 17.11.11 18:54, Sean Bridges написав(ла):
>
> Can you do,
>>
>> Stat stat = zk.setData(somePath, new byte[] {});
>> stat.getVersion();
>>
>> setData(...) will update the version number of the node.
>>
>
> Is it guarantied that two calls won't receive same stat? (E.g. some kind
> of merging of two setData).
>
>
> This is how we use zookeeper for sequence number generation.
>>
>> Sean
>>
>> On Thu, Nov 17, 2011 at 1:59 AM, Vitalii Tymchyshyn<ti...@gmail.com>
>> wrote:
>>
>>> Hello.
>>>
>>> I have a feature request I'd like to discuss: How about a call that will
>>> give next sequential node number without node creation. Main reason is
>>> that
>>> in many cases this would allow retries without leak on reconnect, e.g.:
>>> Now:
>>> 1) create(SEQUENTIAL) - disconnect - retry - and you've just leaked a
>>> node
>>> Can be
>>> 2) incrementCounter() - create(name). And you can make a retry in any
>>> place.
>>> The worst thing that can occur is that you will have a hole in your
>>> numbering
>>>
>>> Another cases this can help are simple counters (no need to create nodes
>>> at
>>> all) or simply different sequential nodes naming schema.
>>>
>>> Best regards, Vitalii Tymchyshyn
>>>
>>>
>
Re: Get sequence number without node creation
Posted by Vitalii Tymchyshyn <ti...@gmail.com>.
17.11.11 18:54, Sean Bridges написав(ла):
> Can you do,
>
> Stat stat = zk.setData(somePath, new byte[] {});
> stat.getVersion();
>
> setData(...) will update the version number of the node.
Is it guarantied that two calls won't receive same stat? (E.g. some kind
of merging of two setData).
> This is how we use zookeeper for sequence number generation.
>
> Sean
>
> On Thu, Nov 17, 2011 at 1:59 AM, Vitalii Tymchyshyn<ti...@gmail.com> wrote:
>> Hello.
>>
>> I have a feature request I'd like to discuss: How about a call that will
>> give next sequential node number without node creation. Main reason is that
>> in many cases this would allow retries without leak on reconnect, e.g.:
>> Now:
>> 1) create(SEQUENTIAL) - disconnect - retry - and you've just leaked a node
>> Can be
>> 2) incrementCounter() - create(name). And you can make a retry in any place.
>> The worst thing that can occur is that you will have a hole in your
>> numbering
>>
>> Another cases this can help are simple counters (no need to create nodes at
>> all) or simply different sequential nodes naming schema.
>>
>> Best regards, Vitalii Tymchyshyn
>>
Re: Get sequence number without node creation
Posted by Sean Bridges <se...@gmail.com>.
Can you do,
Stat stat = zk.setData(somePath, new byte[] {});
stat.getVersion();
setData(...) will update the version number of the node.
This is how we use zookeeper for sequence number generation.
Sean
On Thu, Nov 17, 2011 at 1:59 AM, Vitalii Tymchyshyn <ti...@gmail.com> wrote:
> Hello.
>
> I have a feature request I'd like to discuss: How about a call that will
> give next sequential node number without node creation. Main reason is that
> in many cases this would allow retries without leak on reconnect, e.g.:
> Now:
> 1) create(SEQUENTIAL) - disconnect - retry - and you've just leaked a node
> Can be
> 2) incrementCounter() - create(name). And you can make a retry in any place.
> The worst thing that can occur is that you will have a hole in your
> numbering
>
> Another cases this can help are simple counters (no need to create nodes at
> all) or simply different sequential nodes naming schema.
>
> Best regards, Vitalii Tymchyshyn
>
Re: Get sequence number without node creation
Posted by Benjamin Reed <br...@apache.org>.
you can do a setdata and use the mzxid or version in the stat
structure that gets returned as the sequence id.
ben
On Thu, Nov 17, 2011 at 1:59 AM, Vitalii Tymchyshyn <ti...@gmail.com> wrote:
> Hello.
>
> I have a feature request I'd like to discuss: How about a call that will
> give next sequential node number without node creation. Main reason is that
> in many cases this would allow retries without leak on reconnect, e.g.:
> Now:
> 1) create(SEQUENTIAL) - disconnect - retry - and you've just leaked a node
> Can be
> 2) incrementCounter() - create(name). And you can make a retry in any place.
> The worst thing that can occur is that you will have a hole in your
> numbering
>
> Another cases this can help are simple counters (no need to create nodes at
> all) or simply different sequential nodes naming schema.
>
> Best regards, Vitalii Tymchyshyn
>
Re: Get sequence number without node creation
Posted by Vitalii Tymchyshyn <ti...@gmail.com>.
18.11.11 11:20, Ivan Kelly написав(ла):
> Hi Vitalii,
>
> You could try create(EPHEMERAL | SEQUENTIAL) so that if you loss your session, the znode will disappear.
This will work on connection drop, but not for disconnect/reconnect AFAIK.
Re: Get sequence number without node creation
Posted by Ivan Kelly <iv...@yahoo-inc.com>.
Hi Vitalii,
You could try create(EPHEMERAL | SEQUENTIAL) so that if you loss your session, the znode will disappear.
-Ivan
On 17 Nov 2011, at 10:59, Vitalii Tymchyshyn wrote:
> Hello.
>
> I have a feature request I'd like to discuss: How about a call that will
> give next sequential node number without node creation. Main reason is
> that in many cases this would allow retries without leak on reconnect, e.g.:
> Now:
> 1) create(SEQUENTIAL) - disconnect - retry - and you've just leaked a node
> Can be
> 2) incrementCounter() - create(name). And you can make a retry in any
> place. The worst thing that can occur is that you will have a hole in
> your numbering
>
> Another cases this can help are simple counters (no need to create nodes
> at all) or simply different sequential nodes naming schema.
>
> Best regards, Vitalii Tymchyshyn