cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apache Wiki <>
Subject [Cassandra Wiki] Update of "ThomasBoose/EERD model components to Cassandra Column family's" by ThomasBoose
Date Fri, 10 Dec 2010 14:19:32 GMT
Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Cassandra Wiki" for change notification.

The "ThomasBoose/EERD model components to Cassandra Column family's" page has been changed
by ThomasBoose.


New page:
#format wiki
#language en
= A way to implement EERD components in Cassandra =
== Intro ==
This page describes model tranformations from EERD concepts into Cassandra ColumnFamily concepts.
All input is welcome.

== DBMS layer ==
At several spots in this document you wil find suggestions to implement trivial DBMS functionality
by hand. At this stage, I would suggest to programmers to implement at least 4 tiers when
using cassandra as a backend server. One would be the database layer by cassandra itself,
One would be a tier implementing DBMS rules, another for business rules finishing with an
application tier.

In this DBMS tier functions should be available for keeping data consistend based on datarules
and it would throw exceptions when indexes are changed or orders are given to delete key's
agains DBMS rules.

If this is not yet making sence, read on.

== Indexing ==
In order to add an index to a column, other then the columnfamily's key, we should to create
a second columnfamily. Every insert, which can be either an insert or update in cassandra,
on the original columnfamily we will update the corresponding index.

Think of a columnfamily cf_Person (examples in Python using pycassa)

cf_Person.insert('234', {'name':'Karel','City:'Haarlem'})
cfi_Person_City.insert (Haarlem', {'234':''})

This way a hash will be created containing columns for every person's key that lives in a
specific City. The ColumnFamily architecture of Cassandra can store a unlimited number of
columns for each key. This meens that when deleting a person it reference in the cfi_Person_City
index should be removed first. When updating a person, maybe moving to anothor City, we have
to remove the element from the cfi_Person_City first and then store it with the corresponding
new City.

== Relations ==
=== 1 on 1 ===
Typicly you'll find three kinds of 1 on 1 relations in a relational model. I will address
them one at a time.

==== Equal elements ====
Sometimes all the elements are part of both collections on either side of the relationship.
The reasons these collections are moddeled seperately are most often based on security issues
or functional differences. One solution in a Cassandra database would be the same as you would
implement such a relation in an RDBMS. Simply by sharing the same key in both columnfamilies.
Inserting a key in one of these columnfamily's would insert the same in the other and vise
versa. Updating an existing key in either columnfamily would not result in any change in the
other. Deleting a key from one columnfamily will result in deleting the same key in the other
family as well, providing this would be allowed.

''I'm not sure to what detaillevel security rules can apply in a Cassandra database. At least
I know that one can creat logins per cluster.''

If it gets necessary to use different keys for both collections, sometimes it is not up to
one designer to select both keys, although the number of element are equal and they are related
one on one, in a relational model the designer gets to select either key to insert into the
other collection with an unique and foreign key constraint.

In Cassandra modeling you are forced to either croslink both key's, So you design both key's
foreign in both columnfamily's. Or you create a third columnfamily in which you store both
keys preceded by a token to which columfamily you are refering. Lets focus on the first option.
Say we hand out phones to our employees and we agree that every employee will always have
one phone. and phones that are not used are not stored in our columnfamily. The phone has
a phonenumber as key where the employee has a socialsecurity number. In order to know which
number to dial when looking for employee X and who is calling giving a specific phonenumber
we need to store both keys foreign in both columnfamily's.

-- CF_Employee -----------------------------
|             | name | phone      | salary |
| 123-12-1234 |John  | 0555-123456| 10.000 |
|             | name | phone      | salary |
 | 321-21-4321 |Jane  | 0555-654321| 12.000 |

-- CF_Phone ---------------------------
 |             | employee     | credit |
| 0555-123456 | 123-12-1234  | 10     |
  |             | employee     | credit |
| 0555-654321 | 321-21-4321  | 5      |

Using a static columnname and requiring input in the foreign key fields, checking the existence
of the key in the other columnfamily and processing updates and deletes are all subject to
programming in the DBMS layer. Cassandra itself does not, and probably will not, provide foreign
key logic.  One could imagine an process that makes sure the cross references stay consistend:

cf_Employee.insert('321-21-4321', {'name':'Jane', 'phone':'0555-654321'})

if cf_Phone.multiget('0555-654321', columns='employee') == {}:
   cf_Phone.insert ('0555-654321', {'employee':'321-21-4321'})
   if cf_Phone.get('0555-654321', columns='employee')["Employee"] <> '321-21-4321':
       raise error or delete specified employee<<BR>>
==== Subset elements ====

View raw message