incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Hiller, Dean" <>
Subject Re: Using CQL to insert a column to a row dynamically
Date Mon, 27 May 2013 14:42:52 GMT
Wide rows, dynamic columns are still possible in CQL3.  There are some links here

Also, there are other advantages to noSQL, not just schemaless aspect such as that it can
accept tons of writes and you can scale the writes(you can't do that with an RDBMS).  With
an RDBMS you can typically scale the reads with backups and stuff but there is limits here
too.  There are not limits with noSQL…just double your nodes and get double the read throughput.
This has nothing to do with how much you can store at all.  You maybe are only storing 200G
with an amazing write/read throughput including TONS of deletes to keep it under 200G.

That comes to the next advantage….store huge amounts of data.  If you have 1000 machines
and 300G on each machine, you are storing 300T or 1/3 Petabytes.  Have fun with an RDBMS.

So yes, schemaless is one advantage, throughput is another, total storage room is yet another.
 HA is probably debatable, but in my opinion HA has been another advantage we have seen. 
We have had a hardware outage and no downtime already with cassandra whereas on a previous
project oracle RAC did not really hold up to it's promises.  There may be another advantage
I may be missing as well.

Also, PlayOrm for java client currently uses thrift(astyanax specifically) and so do a ton
of projects right now.  I know PlayOrm is about to upgrade to CQL3 as well so it can do thrift
or CQL3 in the future.


From: Matthew Hillsborough <<>>
Reply-To: "<>" <<>>
Date: Monday, May 27, 2013 8:28 AM
To: "<>" <<>>
Subject: Using CQL to insert a column to a row dynamically

Hi all,

I posted a similar thread on stackoverflow - hope it's not repetitive for anyone here. Looking
for better insight from the community on whether Cassandra is the right tool for me or not.

I am trying to understand some fundamentals in Cassandra, I was under the impression that
one of the advantages a developer can take in designing a data model is by dynamically adding
columns to a row identified by a key. That means I can model my data so that if it makes sense,
a key can be something such as a user_id from a relational database, and I can for example,
create arbitrary amounts of columns that relate to that user.

What I'm not understanding is why there is so much emphasis to predefined columns in CQL examples,
particularly in the CREATE TABLE/COLUMNFAMILY examples:


  empID int,

  deptID int,

  first_name varchar,

  last_name varchar,

  PRIMARY KEY (empID, deptID)


Wouldn't this type of model make more sense to just stuff into a relational database? What
if I don't know my column name until runtime and need to dynamically create it? Do I have
to use ALTER TABLE to add a new column to the row using CQL? The particular app use-case I
have in mind I would just need a key identifier and arbitrary column names where the column
name might include a timestamp+variable_identifier. The whole point is that so I can see have
extremely wide rows at the wonderful performance that Cassandra has to offer. As of right
now, from everything I'm reading in regards to DataStax recommending CQL over Thrift (I think
what I'm describing is possible with Thrift, but correct me if I'm wrong). That means I'd
have to go AGAINST the recommendation to a protocol that's pretty much going to eventually
not be supported.

Is Cassandra the right tool for that? Are the predefined columns in documentation nothing
more than an example? How does one add a dynamic column name with an existing column family/table?
If I'm stuck with static columns, how is this any different than using a relational database
such as postgres or mysql? What I found really powerful about Cassandra is being able to do
something like the following in cassandra-cli which uses Thrift:

SET mycf[id]['arbitrary_column'] = 'foo';

However, doing that in CQL isn't possible. Completely limits the way I was going to model
my data for an application and would have no distinct advantage over a relational database.

Please tell me I'm an idiot and/or am wrong and how I can make this work. It seems Thrift
is the only solution, but I hate going against the recommended protocol.


View raw message