Oh, OK.

> On Apr 15, 2016, at 11:24 AM, Harsha <[email protected]> wrote:
> 
> Hugo,
>        Since this is for rolling upgrade , existing topologies still be
>        using old kafkaSpout with old zookeeper offset storage. 
> Jungtaek,
>           currently kafkaSpout offsets are stored outside of
>           storm.zookeeper.root so we can document saying clear the
>           storm.zookeeper.root node and not touch any other nodes in
>           the zookeeper. 
> 
> -Harsha
> 
> On Fri, Apr 15, 2016, at 11:17 AM, Hugo Da Cruz Louro wrote:
>> Agree with Jungtaek.
>> 
>> With the new KafkaSpout we should not have to worry about preserving
>> offsets, as it’s done by Kafka. We should encourage our customers to move
>> onto the new API. I am going to add more documentation to facilitate the
>> adoption of the new Kafka Spout.
>> 
>> Hugo
>> 
>> 
>>> On Apr 14, 2016, at 6:22 PM, Jungtaek Lim <[email protected]> wrote:
>>> 
>>> Also we need to announce the change of package, and how it will impact to
>>> users.
>>> (There're some questions and some issues on the mailing list and issue, as
>>> you may know.)
>>> 
>>> I think we'd be better to add the following to announce page, with clear
>>> sentences,
>>> 
>>> - cluster rolling upgrade is not supported, so users need to clear any
>>> contents on "storm.local.dir" from all storm nodes and
>>> "storm.zookeeper.root" from zookeeper.
>>> (If we can guide how to preserve the offset information for KafkaSpout that
>>> would be better.)
>>> - Storm 1.0.0 rearranged package name to set the prefix to
>>> 'org.apache.storm' which is not backward compatible.
>>> - Explanation from Abhishek is sufficient for me. (just paste.)
>>> - IMHO I love backward compatibility but we should add the notice that
>>> the hack is "temporary" and users are encouraged to update their topology
>>> code.
>>> 
>>> What do you think?
>>> 
>>> Thanks,
>>> Jungtaek Lim (HeartSaVioR)
>>> 
>>> 2016년 4월 15일 (금) 오전 3:29, Harsha <[email protected]>님이 작성:
>>> 
>>>> Erik,
>>>>       Thanks. Missed that.  I think we should default to that
>>>>       especially to support the rolling upgrade part.
>>>> Thanks,
>>>> Harsha
>>>> 
>>>> On Thu, Apr 14, 2016, at 10:47 AM, Erik Weathers wrote:
>>>>> hey Harsha,
>>>>> 
>>>>> Did you see Abhishek's emailt to the list yesterday?
>>>>> 
>>>>> In the latest version, the class packages have been changed from
>>>>> "backtype.storm"
>>>>>> to "org.apache.storm" so the topology code compiled with older version
>>>>>> won't run on the Storm 1.0.0 just like that. Backward compatibility is
>>>>>> available through following configuration
>>>>>> *client.jartransformer.class:
>>>>>> "org.apache.storm.hack.StormShadeTransformer"*
>>>>>> You need to add the above config in storm installation if you want to
>>>> run
>>>>>> the code compiled with older versions of storm. The config should be
>>>> added
>>>>>> in the machine you use to submit your topologies.
>>>>>> Refer to https://issues.apache.org/jira/browse/STORM-1202 for more
>>>>>> details.
>>>>> 
>>>>> 
>>>>> On Thu, Apr 14, 2016 at 10:35 AM, Harsha <[email protected]> wrote:
>>>>> 
>>>>>> Hi All,
>>>>>>        I tried to lookup if there is any previous discussion around
>>>>>>        this but couldn't find any. Since we moved the package naming
>>>>>>        to org.apache.storm any one upgrading to storm 1.0 have to go
>>>>>>        though the manually renaming their imports to
>>>> org.apache.storm.
>>>>>>         Why not we package backtype.storm and add @deprecated to this
>>>>>>        package and keep it in lib folder for couple of releases and
>>>>>>        have a plan around removing it in future versions.
>>>>>> 
>>>>>> Thanks,
>>>>>> Harsha
>>>>>> 
>>>> 
>> 
> 

Reply via email to