Cassandra

[Bram]

Hi

A follow up on our Cassandra discussion:

Data model elements in pictures:
<http://www.javageneration.com/?p=70>

An extended Cassandra modeling example: <http://arin.me/blog/wtf-is-a-
supercolumn-cassandra-data-model>

A nice and compact overview of Cassandra from the makers:
<http://wiki.apache.org/cassandra/ArchitectureOverview>

Since a lot of people seemed to be interested in practical stuff, many
answers here:
<http://wiki.apache.org/cassandra/Operations>

Then perhaps some discussion/understanding points.

The way I understand it currently (e.g. I could be wrong):

- Indeed all Cassandra nodes know the topology (and thus how to
communicate with all other nodes). One must configure a seed or seeds
on a new node (possibly pointing to a running Cassandra node or
Zookeeper).

- There are several replication strategies, e.g 'simple', rack-aware
and datacenter-aware. In case of the simple one, the N replica's are
stored on the next N nodes on the ring, after the first replica.

- A client connects to one Cassandra node, by using one of the many
different client libs (some do connection pooling, caching, perhaps
also load balancing, fail-over, etc). This Cassandra node knows what
nodes store the key and contacts one, some or all of them (depending
on the client requested consistency) and after receiving the
result(s), forwards the result to the client.

- Perhaps discussing how 'it' actually works, is more interesting than
discussing practical details, since they can be easily found online.
Understanding how it works is harder, so perhaps a better candidate
for a group effort...

- Blahblah


Cheers,
Bram

[Mathijs Homminga]

Hi Bram,

Good stuff. The last point... is that something about Gossiping? ;)

Cheers,
Mathijs