The document provides an overview of MongoDB sharding, including: - Sharding allows horizontal scaling of data by partitioning a database across multiple servers or shards. - The MongoDB sharding architecture consists of shards to hold data, config servers to store metadata, and mongos processes to route requests. - Data is partitioned into chunks based on a shard key and chunks can move between shards as the data distribution changes.

1. MongoDB
Sharding
fundamentals
Antonios
Giannopoulos
Database
Administrator
at
ObjectRocket
by
Rackspace

2. Background
-­‐ 14
years
in
databases
and
system
engineering
-­‐ NoSQL
DBA
@
ObjectRocket
by
Rackspace
-­‐ MongoDB
CerFfied
DBA

3. What
is
sharding?
-­‐ A
mechanism
for
horizontal
scaling
-­‐ Distributes
the
dataset
over
mulFple
servers
(shards)
-­‐ Each
shard
is
an
independent
database
-­‐ All
shards
consists
a
single
logical
database

4. Why
Sharding?
-­‐ Increases
cluster
throughput
–
Read/Write
Scaling
-­‐ Reduces
costs
-­‐
Many
small
servers
VS
one
big
box
-­‐ Eliminates
HW
and
SW
hard
limits

5. MongoDB
Sharding
-­‐ Consists
of
three
elements:
Shards,
Config
Servers
and
Mongos
-­‐ Shards:
Hold
the
cluster
data,
databases,
collecFons,
documents
(Data
nodes)
-­‐ Config
Servers:
Hold
the
cluster
metadata,
map
the
cluster
architecture.
-­‐
Mongos:
Serve
all
drivers
requests.
Route
each
request
to
a
shard
or
shards
(Router
nodes)

6. ApplicaFon
/
Driver
Layer
Mongos01
Mongos02
MongosN
ConfigSrv01
ConfigSrv02
ConfigSrv03
Shard02
ShardN
MongoDB
Sharded
Cluster
Shard01
…
…

7. How
Sharding
works?
-­‐ Range
parFFoning
per
collecFon
(chunks)
-­‐ Shard
key
to
define
chunks
(field(s))
-­‐ Chunks
are
“metadata”
on
the
config
servers
-­‐ Chunks
can
move,
split
and
merge

8. How
Sharding
works?
-­‐
Example
{
"name"
:
"Angelina",
"surname"
:
"Jolie",
"posiFon"
:
"Windows
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Emma",
"surname"
:
"Stone",
"posiFon"
:
"Windows
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Charlize",
"surname"
:
"Theron",
"posiFon"
:
"Linux
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Olivia",
"surname"
:
"Wilde",
"posiFon"
:
"Linux
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Jessica",
"surname"
:
"Alba",
"posiFon"
:
"Sr
Linux
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Scarlef",
"surname"
:
"Johansson",
"posiFon"
:
"Sr
Windows
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Megan",
"surname"
:
"Fox",
"posiFon"
:
"Networks
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Mila",
"surname"
:
"Kunis",
"posiFon"
:
"Sr
Networks
Eng.",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Natalie",
"surname"
:
"Portman",
"posiFon"
:
"Database
Eng",
"phone"
:
"555-­‐5555"
}
{
"name"
:
"Anne",
"surname"
:
"Hathaway",
"posiFon"
:
"Sr
Database
Eng",
"phone"
:
"555-­‐5555"
}
-­‐ CollecFon
employees
for
an
IT
company
-­‐ Shard
key
“posi-on”

9. How
Sharding
works?
-­‐
Example
{
"min"
:
{
"posiFon"
:
{
"$minKey"
:
1
}
},
"max"
:
{
"posiFon"
:
"Database
Eng"
},
"shard"
:
”Shard01"
}
{
"min"
:
{
"posiFon"
:
"Database
Eng"
},
"max"
:
{
"posiFon"
:
"Sr
Database
Eng"
},
"shard"
:
"Shard01"
}
{
"min"
:
{
"posiFon"
:
"Sr
Database
Eng"
},
"max"
:
{
"posiFon"
:
"Windows
Eng."
},
"shard"
:
"Shard02"
}
{
"min"
:
{
"posiFon"
:
"Windows
Eng."
},
"max"
:
{
"posiFon"
:
{
"$maxKey"
:
1
}
},
"shard"
:
"Shard02"
}
-­‐
Lower/upper
bound
and
shard
(server)

10. Choose
a
shard
key
-­‐ High
Cardinality
-­‐ Not
Null
values
-­‐ Immutable
field(s)
-­‐ Not
Monotonically
increased
fields

11. Choose
a
shard
key
-­‐ Even
read/write
distribuFon
-­‐ Even
data
distribuFon
-­‐ Read
targeFng
-­‐ Read
locality

12. Choose
a
shard
key
-­‐ Hashed
shard
keys
for
randomness
-­‐ Compound
shard
keys
for
cardinality
-­‐ Unique
indexes
are
good
-­‐ {_id:”hashed”}
scales
writes

13. LimitaFons
of
Sharding
-­‐ Unique
indexes
–
Just
one…
-­‐ IniFal
collecFon
size
–
Avoid
collecFons
>
256G,
hard
limit
is
a
funcFon
of
key
and
chunk
size
,
for
64MB
chunk/512B
key
is
more
than
1TB
-­‐ Number
of
documents
per
chunk
(250K)

14. LimitaFons
of
Sharding
-­‐ Shard
key
size
<
512
bytes
-­‐ MulFkey,text,
geo
indexes
are
prohibited
-­‐ Some
operaFons
won’t
run
(for
example
group,
db.eval(),
$isolated,
$snapshot,
geoSearch)

15. “Sharding”
–
Other
players
-­‐ ApplicaFon
level
sharding
-­‐ Mysql
(MaxScale,
Fabric,…)
-­‐ Postgres
(pg_shard)
-­‐ ElasFcSearch
(Document
ID
or
rouFng)
-­‐ Cassandra
(Hash-­‐based
-­‐
Ring
topology)

16. Contact
www.objectrocket.com
www.rackspace.co.uk/objectrocket/mongodb
antonios.giannopoulos@rackspace.co.uk
We
are
hiring!
(DevOps,
DBAs
and
more)
hfp://objectrocket.com/careers

17. QuesFons?
Thank
you!!!
MongoDB
Meetup
What's
new
in
MongoDB
3.0
Tuesday,
November
10
,
7:00
pm
@
Harokopio
University