Folks,

agree taking the decorator path is a better option. I have updated -
https://issues.apache.org/jira/browse/FLUME-1097 accordingly.
Will work towards this and post for review.

- Inder

On Sat, Apr 14, 2012 at 5:11 AM, Hari Shreedharan <[email protected]
> wrote:

> I agree with Mike. We can use a decorator mechanism for this. If you would
> like to see this feature, which works in a more generic case. We should not
> be just supporting a "timestamp" header etc. You can file a jira for this,
> to support modification of event headers based on configuration etc.
> Something like this might be useful to even list the hops that the event
> passed through. So I'd rather see a generic mechanism, than a mechanism to
> insert timestamps.
>
> Inder: You are welcome to implement this ;)
>
> Thanks,
> Hari
>
> --
> Hari Shreedharan
>
>
> On Friday, April 13, 2012 at 4:28 PM, Mike Percy wrote:
>
> > On Apr 13, 2012, at 5:49 AM, Inder Pall wrote:
> >
> > > how about supporting something like
> > > "host2.sources.src1.header.timestamp=true" as config.
> > > This overrides time-stamp header on host2->src1(avro source) for all
> events.
> > >
> >
> >
> > Shouldn't it go on the HDFS sink config, so that the HDFS sink /
> BucketWriter uses the current machine time instead of the header time for
> bucketing?
> >
> > If we want to actually modify Event headers inline then we should do it
> via a generic plugin / decorator mechanism instead of one-off features.
> >
> > Regards,
> > Mike
> >
> >
>
>
>


-- 
Thanks,
- Inder
  Tech Platforms @Inmobi
  Linkedin - http://goo.gl/eR4Ub

Reply via email to