cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Zlatanov <...@lifelogs.com>
Subject Re: Apache Cassandra and Net::Cassandra::Easy
Date Tue, 11 May 2010 17:45:06 GMT
On Tue, 11 May 2010 09:40:02 -0700 Scott Doty <scott@corp.sonic.net> wrote: 

SD> I'm trying to wrap my head around Net::Cassandra::Easy, and it's making
SD> me cross-eyed.

SD> My prototype app can be seen here:

SD> http://bito.ponzo.net/Hatchet/

SD> The idea is to index logfiles by various keys, using Cassandra's extreme
SD> write speed to keep up with the millions of lines of logfile we deal
SD> with every day.  Pretty standard stuff, but I seem to be having trouble
SD> getting the software to do its thing.

SD> It says:

SD> _[/home/cass/Hatchet.development/Speed_trial]_(cass@bito)_
SD> $ ./2hk_idx_parser_speed_trial.pl
SD> $VAR1 = 'Can\'t use string ("0") as a SCALAR ref while "strict refs" in
SD> use at
SD> /usr/local/lib/perl5/site_perl/5.10.0/Net/GenThrift/Thrift/BinaryProtocol.pm
SD> line 376.
SD> ';

The latest N::C::Easy will not work with Cassandra 0.6.x, the only
target is SVN trunk.  I can't discover the API version on the server so
there's no way to anticipate such breakage as you see (I suspect it's
due to API mismatch).  The Cassandra developers haven't addressed
https://issues.apache.org/jira/browse/CASSANDRA-972 which can be used to
provide all that information over Thrift as well as in the log files.

Can you give an example of a failing script, with any details on the
server version and keyspace/CF setup?  Without source code I'm only guessing.

Finally, if you're looking for speed, N::C::Easy is not a good fit.  The
Perl implementation of Thrift is significantly slower (5-20x in my
benchmarks) than the equivalent code in Java.  This is outside of
N::C::Easy so there's not much I can do about it (see
https://issues.apache.org/jira/browse/THRIFT-775 which I opened
recently, and which can perhaps help the performance).

Ted


Mime
View raw message