What is Apache Cassandra? | Apache Cassandra Tutorial | Apache Cassandra Introduction | Edureka Video Lecture | Cassandra: Learn and Master - Database Management

[Music]
hello everyone well this session is from
mostly about just rankle I understand
probably I will just give you an eye
level overview of what exactly Apache
Cassandra means and all the relevant
positive use cases that we can solve
that Cassandra and the various possible
scenarios we're in the conventional are
dbmss don't really fit in or they're not
really able to solve the expected needs
so we have to reach out did the
unstructured sources like Apache
Cassandra
don't force won over today's agenda I
will be walking through the conventional
are DBMS rules and principles and the
limitations that we have faced in our
DBMS in recent times and the evolution
of big data or how the big data has been
evolving the rise and dominance of high
volume and extensive throughput data and
the interactions words the unstructured
sources that we primarily reference
words in the West or there are the
various possible flavors of no SQL or
the various characteristics of a know
SQL database or ours the various key
elements of the various flavors of no
SQL obviously primarily differs or all
and then probably will be going through
a few of the case studies so let me try
to down Lord
Cassandra once more on my local machine
- Wow we can go through the case studies
not only if you have a case studies on
my machine now but the conventional our
DBMS obviously our DBMS is happening in
the industry for a very long time
because of the security and the
specifications the level of features
that they provide we always can rely for
a good amount of input and output
operation the transactional managing of
the data is being preserved or had the
data is actually being processed based
on the various implementations for
example if we go with Oracle the scope
of the broad spectrum of Oracle might be
huge and immense as compared to an open
source or previa like my
QL or even post playback and then if we
talk about the SQL servers every odbms
has its own flavors yes on the base
level the principles and the structures
the architecture and itself on the code
does not really change it always remains
the same
so there's always the various flavors
with respect towards managing the memory
how optimize memory usage with therapy
and the source what it will be the
constructive ways of processing the
information or the data that is there
and and the distribution of the data I
mean and then the sophistication after
which the data can be extended or 1/2
the data would actually really be able
to put that off in here so now moving
one this line I was saying that the
structure of the data that we have in
the our DBMS so there's always been
extensive amount of sophistication and
the biggest part of that is probably
have the data is being preserved in the
our DBMS predominantly in a relational
structure and how we can distribute and
how we can apply that off in various
levels of environments and ecosystems
that has always been an edited bondage
in ER DBMS obviously as I was saying
that there are various flavors of our
DBMS you are proprietary fewer
commercial editions based on how the
industry has evolved over the past two
or three or decades now how much the
challenge the respect towards the
industry how the industry has moved
along now not since there's a lot of
extensive amounts of this communication
that has been going on across in the
industry now there are various levels of
devices that we're trying to automate
simplify our processes and they have
various structures or various formats in
which the data is being generated from
those devices so eventually the growing
needs of managing or maintaining such a
high volume of data is has been going on
and very high and a very rapid pace
these days now we've been looking around
for this problem to solve
extensive amount of data that is not
really a conventional stored in X
structure in an organized manner there
might be a lot of data for example if
you see how Facebook is being able to
maintain so much a high volume update
having you have more than billion odd
user
and how every user everything believes
that is trying to you make use up that
platform is different as compared to the
others
the remaining one so it gets really
difficult to actually manage that volume
of data so now that's the reason why the
unstructured source of data has been
increasing immensely at a very very high
rate and especially in the past decade
there have been data bytes of data that
has been piled up across the industry
so now as computer words from the
conventional structure source since the
industry since we're trying to go in an
automated environment not a Medeco
system where we're trying to monetize
the data more specific towards a
specific entity as compared to
collaborating everything at one single
source like a conventional source that
we used to do in the legacy systems that
approach had changed so the dominant the
data is more loosely couple and more
unstructured data now with respect
towards how the presumption has been
going on based on that that how
eventually in the next couple of years
they will predominantly be more data on
the unstructured side probably around
seventy to eighty percent of data I
would eventually if I end up being in
the unstructured format as compared to
the conventional structure and
relational format so that means there's
more volume of data that is actually
coming up that is immense this structure
is not really specified so there is no
specific format or pattern that will
really follow eventually to preserve
that data so based on the kind of data
we eventually end up with an extensive
amount of data that needs to be managed
as per the pattern or as per the style
of the data over the comparison in the
legacy systems or in the conventional
our DBMS system wherein we use to have a
data with respect towards a specific
structure or specific schema what we
call as no the predominant high volume
of data even though there might be a lot
of useful and unuseful data that would
be eventually available or in that
specific zone or in that specific area
that is mostly referred to as Big Data
since there's high volume of data that
actually gets generated or that gets
created due to various possible
scenarios for example if we talk about
YouTube if we talk about Twitter
LinkedIn
Facebook most of remaining other social
media obviously social media or the
e-commerce or in the IOT domain there's
always high volume of data that gets
generated and the structure of the data
always keeps on varying there's no
specific structure format that
eventually gets specified or that based
on which the data gets aligned towards
that now eventually this is not really a
standard on its own however
conventionally most of big data is
eventually being represented by the PI
famous Lee's volume variety value
velocity and velocity they're in volume
obviously means there is a high amount
of volume that needs to be processed
probably in smaller chunks are smaller
components where in the variety of that
means the flavor of the data our friends
always varying there's no consistent
structure of the data that we'll be
expecting from a source so there will
always be these flavors they might be
data in the form of media files in my
detail and the form of JSON XML there
might be data in the form of blog files
mails and so on so and all of that data
will be useful to us and that needs to
be processed eventually now with respect
to the velocity the concept of big data
eventually is that there is always high
volume of data and at the rate at which
that data gets generated or that data is
created is always at a very high speed
that's what the velocity eventually
refers to and then I don't know since
there is always high volume of data and
there is an extensive bulk amount of
data that gets generated but not
necessary you will always find every
single item has to be relevant to your
need or an use case there might be a lot
of things that need to be fine-grained
that need to be cleansed off from the
actual high volume data that has got
generated so there might be only
specific elements or entities that are
more relevant and that are more suitable
or is there a more logical and more
specific towards use case that is what
we refer to as the meaningful data or
valuable data as for our case now the
veracity is primarily more towards what
is the probability of consistency in the
data so now since
we don't just like I said that there's
no specific consistency there might be
one single source I really from Facebook
if we pull out data from Facebook it
might vary with respect to towards the
users and the kind of sources that
Facebook is communicating to as compared
to Twitter so the data that gets
generated by Twitter
or did I get generated by Facebook will
eventually be different will not really
be symmetric in nature so there might be
possibility that if we imagine a try to
store deserve the data from both the
sources in a simple story structure or
persistence store so we might not really
have a consistency in the data because
since there's always possibility that we
don't know what kind of data will be
sent from Facebook at any point of time
the probability is always very
inconsistent now the industry has gone
very stable in the computer science
environment so now the probability of
data leak has grown now so the data
mining the data filtering the cleaning
of data so that the data science
ecosystem has grown into a very broad um
spectrum now so you know always various
extensive mound of possible big data
sources that are available in the
industry as of now and eventually we
might be part of either the generation
source or the maintaining of the data so
now in both of the scenarios the source
of the data are always divided in parts
so we eventually need to maintain till
those amount of sources of data now if
we compare if we try to solve before the
existence or the evolution of the new
sql's that have boomed the industry in
the recent times people actually try to
solve up the problems to the ID DMS
obviously there was a breaking limit
where we're not really able to maintain
or preserve that much amount of data
basically cost was a very big factor
because we're not really the licensing
cost of most of the commercial editions
in the RDBMS store is very very high and
then with respect to the high volume of
data the scalability of the system and
the performance obviously that gets a
very big challenge so that means now in
case if we want to have an optimal
performance we always need to go for a
very extensive and very very expensive
and highly configurable of horizontal
scaling what
scaling itself is has its own
limitations and then maintaining the
unstructured sources the structure of
the data in our DBMS is always a big
challenge because there is no specific
structure based on which how the our
DBMS really expects that off so there
are a lot of big challenges there and
our DBMS has failed to sponsor the high
volume or the big data environment so
now for example if you take these are
few real-time it's - like an Instagram
the amount of data actually grew up
extensively at a very rapid pace and
then in order to give an extensive
amount of performance putting a lot of
data in the in-memory store was not
really a feasible option or basically
because it was not really cost-effective
and then eventually it was actually
really affecting the hardware
performance in there and then the
limitation the day of IO operations with
respect to the RDP MSL is always a
performance glitch in there because
they're always limitation there is no
specific concept of size scalability nor
dbmss and then managing that much amount
of volume of data and then understanding
the multiple points of failure that's
always challenging in their notes for
the respect word scalability from the
conventional no SQL stores almost every
new SQL store has that concept of
cost-effectiveness because primarily
most of the popular noise schools are
eventually free of cost
they don't really charge a dime except
for a few of the proprietary ones and
then the processing speed off the noise
fuels to manage the unstructured source
at a very high volume is very immense
really very amazing and the availability
of this system or the echo system is
very very quick because most of the
noise will eventually support the
distributed environment so there's no
single point of failure available that
is being made available in conventional
no SQL store and same applies to Apache
Cassandra
so that's where the new ice tools
actually come in picture and they really
saw us today in there so for no sql's
they're sophisticated mechanisms for
sharding for distributed data you can
scale up the ecosystems but it
and Horizonte at any point of time there
is no single point of failure in there
and then there is no specific
restriction on the storage the format or
the structure of the data that we are
trying to preserve in the system it's
completely agnostic with respect to
works the data that we are trying to
preserve in there now the respect
towards the noise tools there's a wide
range of noise fuel stores that are
available there in the industry and
predominantly there are being
categorized into based on the structure
in which they stores the data we have
the column stores the column based words
that is we have Cassandra and HP's
document stores that store data in the
form of documents or JSON we have
CouchDB and Mongo for the key value
stores we have dynamo and Redis for the
graph databases that have a few of the
qualities of the transactional stores as
well for a more healthy migration when
and when we have relations if the graphs
are more healthier than asking people to
the remaining of the trees we have neo4j
orient tied in and so on and so go now
with respect to the performance
obviously Cassandra stands by far one of
the most optimized and more
sophisticated in the industry even
though
dynamodb and Mongo stand very close to
Cassandra however the best part about
Cassandra is that Cassandra has a very
good concept of giving a consistent
performance with respect towards reads
and writes
now digging more deeper into Apache
Cassandra just like I was saying that
sandra is an open-source system that was
predominantly being developed by a few
of the developers at Facebook and then
it was put under the Apache umbrella
then from the incubation it was moved
into the tier one project and since then
it has been a very prominent store and
Industry has accepted the structure of
Cassandra based on the performance and
the credibility that Cassandra has been
providing to its users based on that
industry has accepted cassandra it's
actually another very high paced so
there are a lot of production based
deployments that have been running on
top of cassandra for around a decade or
so even though Cassandra was mostly it
in her small things from Google's
BigTable and there has been a lot of
comparison with respect worse BigTable
Han Apache Cassandra
now with this pictures for a comparison
per se with respect to and our DBMS and
Cassandra and the architecture Cassandra
supports primarily works on gossip
protocols so there's a peer-to-peer
communication that happens there's no
single point of failure because that
doesn't really work on the concept of
masters and then what the availability
and obviously in Cassandra Cassandra
always has a very highly available and
scalable system where in the volume of
data and the structure of data is
doesn't really have any boundaries they
can always scale up and the availability
of the system can always go up because
cassandra has a very good concept of
snitches that support the adaptogen on
based on which hardware's the cassandra
is actually very applied however in case
of an odbms conventionally really are
more tight boots the master/slave
concepts so there needs to be a master
that takes care of the responsibilities
and the slave is responsible eventually
to handle and cater the requests that
are being triggered by the master itself
and obviously there's always the only
structured sorts of data that can be
preserved in our DBMS
that is not really a restriction or
limitation off Cassandra
cassandra has a really good support for
the data center so I mean conventionally
I mean in all the times I mean this
makes more sense
environment or in a distributed
environment as compute to have single
source or single set of deployments or
in-house deployments it's like we used
to perform that an AR DBMS and we have a
datacenter on and we've managed
everything in there and then the
challenge is that if the data center is
down so we lose the data or we have to
wait for the data center to get restored
in order to buy back the data that is
available in there and until that the
response from the system might be
sporadic or we might not get any
response so this loop consistency in
depth however with respect to multi data
center catering so there's always
support that you can have
Cassandra applied across multiple data
centers that means if there is a data
center in East Zone one probably in
Arizona or Oregon in US and there's a
lottery the Center in California and
then there's probably our data center in
Europe probably in Germany or London so
we will still be able to respond and
even if one of the data center it's down
we will not lose the data that data will
at least have specific backup or
response at one of the guitars the
remaining other data centers will have a
copy off the pause data center now with
respect to in memory almost everyone has
in-memory support whether that's odbms
or cassandra same applies to these
search supports but there are a few
sophisticated supports that you can do
and the performance of the searches
obviously will be faster on Cassandra as
compared to the our DBMS and then
eventually not every IDB emissary's
provides a very good support for it
mostly the commercial editions or the
propriety tools that have a good support
on these search systems talking about
the extensive features of Cassandra
obviously with respect towards the right
there's always consistency and they're a
very effective there's a very good
mechanism for fault tolerance and high
availability and scalability of the
system and then the distribution of the
data based on the partitioning key and
how did that gets distributed across and
then how the sharding actually works in
Cassandra that's very very quick and how
exactly it doesn't really affect the
system even if there are millions of
transaction that are happening which
they're in Cassandra's so we will still
be able to achieve a very very good and
optimal and consistent performance
Cassandra gives very good flavor of its
own to traverse or past communicate with
Cassandra in its own query language the
Cassandra's four little tiny CQL let's
ping build on top of drift protocol so
but it has been written more reliant
words for people who are coming from a
conventional background for my DBMS so
the transition becomes more simpler and
easier obviously industry one of the
biggest consumers of cassandra as they
clicks netflix has been using the sun
rod with high volume and millions of
millions of transactions are being
stored in their this comcast who has
been using cassandra for a pretty long
time now Spotify users to maintain their
products and playlists in Cassandra as
well and then there are a lot of other
users for that's NASA apps Mantic Cisco
it will be IBM if they did SoundCloud
Twitter and so on and so we're obviously
Facebook is there is because one of the
bigger users as well now in order to
understand what Cassandra really brings
on the table we need to really we should
be pointed with popular terminologies
that are there in Cassandra Cassandra
conventionally on top everything is
being represented or everything is being
put under the umbrella of clusters that
is the high-level container that
preserves everything else in its or an
na cluster can be spread across multiple
lords and in a distributed environment
then eventually it's the key space key
space is very very identical to a
database from an RDBMS perspective there
in all the data is actually being
preserved inside the key spaces
colostrum can't have a bunch of p spaces
and then that's how the tree actually
moves on and then we have the column
family koala family is eventually very
identical towards the table even if you
traverse you see QL you will see that
there is a lot of identification or
similarity between the table and call a
family even though I mean this
reservation are for color families
obviously quite different as compared to
table and I DBMS but the terminology has
been made such and looking
seal has been made such so that we're
able to correlate in between them and
then we have the corresponding tables or
eventually we might just have the set
columns that represent or that build up
with respect towards these column family
impair now for the clustered is
eventually that is responsible to build
a proper ring or a cluster in between so
as a communication with respect towards
the multiple lowers across the cluster
is being done in a more efficient and
more simplified way so that the
communication is more stable in its own
Bay and its nature and then maintaining
following the principles after
replication and starting how data gets
distributed across the various nodes and
then how to maintain the replication in
there as well then when we come to key
spaces we got the key spaces key space
is eventually similar to an identical
database in our DBMS where it might
preserve it might have a huge bunch of
smaller families eventually in its own
that will have excavate a set of columns
in there and then my key space has few
characteristics it has name and then it
might have tragedy based on which the
key space has been created based on
which cassandra decides how do I need to
distribute the data across the various
nodes how do I need to perform partition
and clustering for creating the key
space and become on is very simple you
can specify the create key space and
widths we can specify the replication
factor that is associated in there so we
can have their application factor so how
many replication do we want for this key
space and the data inside it and the
durability of the rights how durable are
the rights in there and then we have a
replication strategy for the replication
strategy cassandra echo comes up and get
two different strategies the simple
strategy and that the simple strategy
mostly talks about how the ring really
works and how we design the key space
and its inner elements inside it and the
network topology is probably our best
ear more sophisticated but we're and
we're more into a highly and more
tunable system in a security environment
if we want to do more specifications in
a distributed environment that's where
the network topology actually really
makes if we want to be more specific
towards which racks are we
talking about so that's where the
network topology will actually be
talking about Oh in case of simple
strategy it primarily depends on it
relies on one single data center and
everything gets preserved in there so
eventually it makes more sense to have
network topology strategy in your system
or in your configuration when we're
working in across multiple data centers
and by that way we'll eventually be
saving more data and we'll be able to
manage our data more optimally for the
column family just like I said it's
really identical to would like a table
that we have eventually it's the bunch
of words that preserved it and the form
of columns and then obviously that Paula
might have a key and value the key would
be the name of the column and the value
would be the value of that poem so now
every row will be different even though
there are a bunch of rows that get
associated inside the same column family
our every rule would be different not
necessarily that you will have one row
that will be identical out that will
represent the role in there's every rule
will be represented by a unique
identifiers that we refer to as the row
key and then the room key eventually may
or may not be a part of the primary key
that we have the unique identifier based
on which Cassandra decides that okay
this is the amount of data that will
actually be going in this specific store
or not see too old table eventually it's
the same color family now in order to
create this is simply the syntax of
creating the cq3 book how eventually we
end up creating 3cq our table and how
will that we look like okay now before
actually going on to the use cases let's
try to do few examples they see let's
try two
we start Cassandra
okay see Cassandra so I will let's start
Cassandra in here and let's start CQL in
the other window Lords
let me maximize this so we have people
we have a chica Sondra then cql I search
strike log n so we use describe key
spaces so when you say describe key
spaces we can use either describe or we
can use this de señor
it's the same thing so we can use either
of them in order to use a key space we
say use system now if we say describe
these space system it will give us the
information about the key space that is
there and the tables that are associated
with this now let's try to create
key space so let's see it will recur
Sandra we create
okay sorry we have to use vacation
okay so we have created the key space
with a create a space name of the key
space with the class the replication
strategy and the replication factor and
the durability so now if you say
describe these space a to Rekha
Cassandra and gives us the information
about it so now we can say use a - Rekha
Cassandra okay now we say create so we
say create let me keep it in uppercase
for simplicity we say create table we
say test we create for example ID
time uid we have new text and we have
email text then we have primary key
and we have 80 so we have create the
table name test the three fields we have
ID we have name and we have email we're
all live there in the cluster we go to
the key space and then we clear the
table name test
there's a default cluster that is being
created by the Sun on its own that is
the DES class or if you see up a table
when we started it says connected to des
cluster so if we want to change we have
to change the confirmation and the
casandra's config file that's Cassandra
Dorothy mo file for now let's not do
that so let's create art in here
clear that off stay sample time UID so
we can copy that it's for being the same
let's tutor
okay so now we've entered the time uit
now let's see so now we say select star
from test here it is if you see there is
a new record that got created inside in
here so we created key space then we
created column family inside that then
we defined the time the key for that
that's how the work he would be created
and would based on which the partition
algorithm in Cassandra will start
working and then on top of after that we
inserted data inside that and then we
retrieve that data in there similar to
the basic other simple plan operations
that we eventually do so similar to that
we are able to perform the operations in
there was trying to take a kid study
let's take the example so let's take a
look at the case study
you're working in the shop or calm your
company sells products customers only at
some specific locations you initially
started with half a million products you
now need to store all available in any
data store
you're not going with MySQL you've
created a primary key and some secondary
indexes on few of the columns and
strongly using it for a container for
abuse with them our name description
chord or DM RP and so on there might be
sure that feels as well called a catalog
can be huge in assume it's dating
frequently a few years you got funding
from Kim Jong to expand your business in
different countries the product got
increased from 1 million two hundred
meter you're facing a challenge the
products were different and have
different attributes stories for not
going the kisana which is horizontally
scalable and provides high availability
keeping all things in mind we need to
design the schema force the product
catalog with the below bits the
attributes are primary key partitioning
P secondary indexes table details any
views that you're required and we need
to keep in mind for our details
by-product chord we need to have
products by sellers specific category
where products by same category belong
to same brand products by brand and
category products by seller core and
store now let's take let's create key
Spears
use a direcor we created key space time
if you're a car and then we create a
table with primary key affairs for a
core
then we clear when we can spark or then
we insert the attributes in there so
let's see
so we say select select star from so we
have entered an attributes with for
accord bran category creation time
description MRP triose for our name
quantity and cell accord so now we have
added we try to select the attribute
when we try to select or attributes and
then it gives us the attributes namaste
now if we try to select the specific
attribute with respect using the primary
key then we will be able to achieve the
primary key now if they are actually
trying to select star from products
their brand
it's saying that hey this is not
possible since none of the query
parameters have a primary key with them
associated while in the where clause nor
is there any attributes that specifies F
filters or that is already being
specified so now in order to use a noun
to receive the data from Cassandra
without using the primary key so we need
to allow filtering in there so even
though it makes it will make the query
slower and to take Cassandra more time
to filter that off because then the
Cassandra needs to go for a random
search and then it will take assign raw
some time to go through that
now the alternate or the solve that
challenge this like I was saying the
best way is to create indexes let's go
and create Linux
now if we go back in here
now when we hit without enabling the
filtering it is still able to receive
that because since the data is being
preserved using the index based on the
index so now index really gets applied
on that specific column so the data that
we are trying to search in that column
would be kept in specific different
stored in there similarly we can have
multiple indexes based on
knowing guess if we want to be the
indexes let's write you the list of
indexes that we have here we're trying
to view the column name the index name
the index attributes and from system
dots scheme underscore columns right for
the key space head Eureka and the column
family this is other way in case we want
to create a different table to maintain
data Air Force the sample data in there
so this is another way if you don't want
to use indexes we can push the data in
there so create the table here in new
table products brands and that has four
according brand and then we insert the
data in there
let's switch now
now this is other way using a similar
approaches like we use it in our DBMS is
we have a relational tape we have the
actual primary table then we have the
secondary relational tables where we
maintain only the variation or the
foreign key is Dawson there is no part
of the foreign key so there will not be
any specific ruling on the next table
that is the Forex brand table so we can
just preserve the brands and there
however there is no possibility of
consistency to impose the specific
restriction may be the consistency now
for that what we can do is we can make
use of the materialized views
materialized views are eventually just
to specify a higher level pictures of
the data so now in case if we want to
specify
so now we first we created the Mitchell
ice view 4x4 brand we're selecting data
from for xb1 and the respect towards the
brand so based on the brand where the
brand is not feeling now so we have a
specific brand in there these are
similar to views just like we had the
view so that means the pre-compiled
curious that we used to have in SQL
similar to that we have materialized
views that was introduced in cqm fear so
since then it has been part of cql even
though it's not something that we use
you will see the usage or the
expectation in the other sources however
but still this is one of the
possibilities our base
using Cassandra can have the parts by
brand with respect to deep ran and we're
gonna have you dated inspect words parts
very that don't really have a bra come
to you either we keep all the products
or we can achieve the board a products
based on a specific brand in here this
is probably a little brief of how we
approach Cassandra from the basic set of
queries using cql how we configure cql
we just need to download this zip off
Cassandra and start exploring the
Cassandra you can easily explore you can
download that off you can put the bin
folder start Cassandra from your command
prompt and then run CTL and on Windows
it comes as an installer or you can
download the Cassandra from rpm on App
gets well that's predominantly an
overview of how Cassandra really looks
like and what are the various aspects
and were ours what else the basic brief
overview of Cassandra based on which we
can say that that's how Cassandra really
looks like so I thank you again for
joining in and for watching this video
in case you have any doubts feel free to
reach out to us at any point of time
alright thank you and have a great time
ahead
why
I hope you have enjoyed listening to
this video please be kind enough to like
it and you can comment any of your
doubts and queries and we will reply
them at the earliest do look out for
more videos in our playlist and
subscribe to any Rekha channel to learn
more happy learning