hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Purtell <apurt...@apache.org>
Subject Re: Structure of values map in HColumnDescriptor.
Date Sat, 11 Apr 2009 22:29:29 GMT

It was I who did it that way.

-1 on reintroducing Text.

No need to store the values as strings. We can store
ints as varint representation in byte[], etc. I used
strings only so printing the map from the shell gives 
reasonable human readable representation, with the
expectation that the values will be rarely consulted
beyond startup and region deployment times. If this
consideration has become inconvenient for other uses
and those uses outweigh the readability consideration,
+1 to use Bytes and/or Writables to marshall and
unmarshall other types avoiding transitions through
string objects.

   - Andy

> From: Erik Holstad <erikholstad@gmail.com>
> Subject: Structure of values map in HColumnDescriptor.
> To: hbase-dev@hadoop.apache.org
> Date: Monday, April 6, 2009, 6:12 PM
> Today we have a values map in HColumnDescriptor that maps
> strings of names to ints, booleans and strings.
> For a int value you make a string out of it, then you do
> getBytes on that string to get the bytes and finally new
> ImmutableBytesWritable to store it in the map.
> I'm not a big fan of the Text format and I'm happy
> that we are trying to get rid of it in as many places as
> possible but I think that this is one of the cases where
> is would actually make sense to keep them, so we don't
> have to create all these new objects just to write them
> and have something like a map<Text, Text> or map<Text,
> IntWritable>.
> If choosing the last option we need to convert the one
> place where strings are used for the values,
> compressiontype, to an int.
> Regards Erik


View raw message