Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Flume >> mail # dev >> ElasticSearchSink: should we combined the proposed interfaces for event serialization and id assignment

Copy link to this message
Re: ElasticSearchSink: should we combined the proposed interfaces for event serialization and id assignment
I added you to CC but I would also recommend subscribing to the dev list
due to how the list headers are configured.

This is a rough situation. I am loathe to break API compatibility but at
the same time I don't know much about ES and feel I need to find some time
to invest in understanding the ES concepts and how Flume is interacting
with them in this sink.

Without thinking about it very hard yet, I'd ask if we can just add a new
interface that doesn't suck and maybe is more extensible without breaking
the old one.

Also wondering if other devs have thoughts on this.

On Thu, Apr 18, 2013 at 11:57 AM, Edward Sargisson <[EMAIL PROTECTED]> wrote:

> Hi all,
> (I read this list in digest mode; would you mind ccing me on any replies?)
> I've got two patches progressing through Jira (FLUME-1782, FLUME-1972).
> -1782 fixes a defect where the wrong timestamp field and elasticsearch
> index name are used. -1972 adds an interface which users can implement to
> assign an id instead of letting elasticsearch randomly assign on.
> The question to discuss: should we(I) combine those interfaces and just
> have a single interface.
> Mike Percy has kindly reviewed FLUME-1782 and the knock-on effect of his
> comments require that the ElasticSearchEventSerializer interface be changed
> - and thus this becomes a breaking change. I had been attempting to avoid
> that and this is why -1972 has a new interface.
> If we're going to break the interface then maybe we should go all the way
> and put the new id provider functionality on to it as well? We could also
> rename it to ElasticsearchEventSerializer (lower case s on search) to match
> the way the maintainer of elasticsearch spells it.
> The new interface would be:
> public interface ElasticSearchEventSerializer extends Configurable,
>     ConfigurableComponent {
>   static final Charset charset = Charset.defaultCharset();
>   XContentBuilder getContentBuilder(Event event, Long timestampOverride)
> throws IOException;
>   String getId(Event event)
> }
> The timestampOverride would only be non-null if there was no timestamp
> header.
> Thoughts?
> Cheers,
> Edward