Home | About | Sematext search-lucene.com search-hadoop.com
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB
 Search Hadoop and all its subprojects:

Switch to Threaded View
Zookeeper >> mail # user >> Watching children updates


Copy link to this message
-
Re: Watching children updates
ZOOKEEPER-1416 is another way of solving the problem of watching a sub-tree

--
Thawan Kooburat

On 7/17/13 10:50 PM, "kishore g" <[EMAIL PROTECTED]> wrote:

>Good idea beware of failure scenarios.
>
>lets say you do
>#1. parent.setData #2 child.setData
>
>your watch might be triggered and when you read the children child may not
>be changed yet. so u will miss the child data change
>
>other option is
>#1 child.setData #2 parent.setData
>if you fail after setting the child data, there wont be any watch
>triggered.
>
>Only option is probably doing child.setData and parent.setData in the same
>transaction and probably the multitransaction can be helpful. But if you
>are updating children often the multitransaction might fail quite
>frequently because of concurrent updates on the parent
>
>I think its better for zk to solve this problem the right way. Instead of
>solving the problem of large number of watches in zk, it might be a better
>idea to add the concept of watching childDataChange in zookeeper. I dont
>think the implementation will be any more complicated than triggering
>watches for child creation.
>
>Thoughts ?
>
>
>
>
>
>On Wed, Jul 17, 2013 at 6:23 PM, Benjamin Jaton
><[EMAIL PROTECTED]>wrote:
>
>> Thanks, the JIRA is very interesting.
>>
>> I am contemplating the possibility to make a parent.setData(..) call to
>> notify of a change in one of the children and put a watcher on
>> parent.getData().
>> I would then iterate through the children and compare the
>>stat.getVersion()
>> to figure out which have changed.
>>
>>
>> On Wed, Jul 17, 2013 at 5:30 PM, Flavio Junqueira <[EMAIL PROTECTED]
>> >wrote:
>>
>> > Hi Ben,
>> >
>> > Your description seems right to me, assuming that when you refer to
>> > changes to the children, you mean setting the data of the children.
>> >
>> > You may also want to have a look here if you haven't done it yet:
>> >
>> >
>> >
>>
>>http://zookeeper.apache.org/doc/r3.4.5/zookeeperProgrammers.html#ch_zkWat
>>ches
>> >
>> > About performance impact, you may want to have a look at the
>>discussion
>> > here:
>> >
>> >         https://issues.apache.org/jira/browse/ZOOKEEPER-1177
>> >
>> > -Flavio
>> >
>> > On Jul 18, 2013, at 12:25 AM, Benjamin Jaton <[EMAIL PROTECTED]>
>> > wrote:
>> >
>> > > Hello,
>> > >
>> > > I would like to be notified of a change on any of the children of a
>> znode
>> > > (could be a lot of them).
>> > >
>> > > AFAIK to achieve this we need to set a watcher on each child and
>>then
>> > also
>> > > watch the parent to be aware of children addition/deletion and
>> add/remove
>> > > watchers accordingly. Is that correct?
>> > >
>> > > What is the performance impact of having for example thousands of
>> > watchers?
>> > >
>> > > Thanks,
>> > > Ben
>> >
>> >
>>
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB