hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From lars hofhansl <la...@apache.org>
Subject Re: [ANNOUNCE] Secondary Index in HBase - from Huawei
Date Wed, 14 Aug 2013 22:35:13 GMT

1. highly selective indexes + point gets -> global inverted index tables
2. less selective indexes + queries returning many rows -> "local" indexes, such as the
Huawei solution.

Of course it's not quite that black and white. Global indexes that serve index covered queries
(where the query can be answered from the index alone) would also work in many cases of non-selective

In the end it is quite simple (IMHO):
If a query retrieves data from only a single region, you want to able to hone into that region
quickly, via a piece of global information.
If on the other hand a query returns data from many regions, you're better off handling the
filtering locally.

Just my $0.02.

-- Lars

 From: Andrew Purtell <apurtell@apache.org>
To: "dev@hbase.apache.org" <dev@hbase.apache.org> 
Sent: Wednesday, August 14, 2013 1:52 PM
Subject: Re: [ANNOUNCE] Secondary Index in HBase - from Huawei

On Wed, Aug 14, 2013 at 8:45 AM, Michael Segel <michael_segel@hotmail.com>wrote:

> This isn't too bad if you're doing a simple query against one index. You
> can do the work by RS and then join the results from all RS.
> However… what happens if you have two indexes and your result set is going
> to be the intersection of the indexes?
> Or you're going to do a join between two tables using the indexes to limit
> the result set?
> Now your design breaks down quickly.

You may have just described their design assumptions.

I'm not endorsing this per se, but suggesting it is not a good idea on
account it can't live up to the requirements of a pretty particular
strawman seems a step too far.

Maybe someone from Huawei can talk a bit here about successful use cases?

> You could also look at Lucene which we did a PoC a few years back.

A certain large technology company has an HBase full text index built on
Lucene that might be offered as a contribution at some point. From what I
know of it, there are a different set of tradeoffs and it certainly won't
work for everyone, and not because the people working on it were not smart
enough to find a silver bullet.

Best regards,

   - Andy

Problems worthy of attack prove their worth by hitting back. - Piet Hein
(via Tom White)
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message