Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Scaling MongoDB | Sergey Gavruk
Search
Minsk MongoDB User Group
October 04, 2012
Programming
2
170
Scaling MongoDB | Sergey Gavruk
Sergey Gavruk
Meetup #7
Minsk MongoDB User Group
October 04, 2012
Tweet
Share
More Decks by Minsk MongoDB User Group
See All by Minsk MongoDB User Group
MongoDB by Chef | Yauhen Artsiukhou
bymongo
0
120
MongoDB at IronMQ | Alexander Kolesen
bymongo
0
840
Event sourcing + CQRS + MongoDB | Alex Shkor
bymongo
1
640
How it works. Indexes | Kirill Duborenko
bymongo
5
270
Aggregation Framework | Mikhail Burtylev
bymongo
1
92
MongoDB 2.2: Release update + Roadmap | Alvin Richards
bymongo
1
92
Meetup#6 Intro | Alex Litvinok
bymongo
1
44
Deploying MongoDB on Amazon WS | Michael Karpitsky
bymongo
2
110
About the problem of DBMS choice & what to do if you have gone the wrong way | Roman Bugaev
bymongo
3
120
Other Decks in Programming
See All in Programming
Jaspr Dart Web Framework 박제창 @Devfest 2024
itsmedreamwalker
0
150
chibiccをCILに移植した結果 (NGK2025S版)
kekyo
PRO
0
130
オニオンアーキテクチャを使って、 Unityと.NETでコードを共有する
soi013
0
370
Azure AI Foundryのご紹介
qt_luigi
1
200
月刊 競技プログラミングをお仕事に役立てるには
terryu16
1
1.2k
Итераторы в Go 1.23: зачем они нужны, как использовать, и насколько они быстрые?
lamodatech
0
1.4k
ドメインイベント増えすぎ問題
h0r15h0
2
560
Beyond ORM
77web
11
1.6k
カンファレンス動画鑑賞会のススメ / Osaka.swift #1
hironytic
0
170
LLM Supervised Fine-tuningの理論と実践
datanalyticslabo
8
1.9k
ASP.NET Core の OpenAPIサポート
h455h1
0
120
快速入門可觀測性
blueswen
0
500
Featured
See All Featured
Facilitating Awesome Meetings
lara
51
6.2k
4 Signs Your Business is Dying
shpigford
182
22k
The Art of Programming - Codeland 2020
erikaheidi
53
13k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
33
2.7k
Building a Scalable Design System with Sketch
lauravandoore
460
33k
Why Our Code Smells
bkeepers
PRO
335
57k
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
49
2.2k
Writing Fast Ruby
sferik
628
61k
Become a Pro
speakerdeck
PRO
26
5.1k
Why You Should Never Use an ORM
jnunemaker
PRO
54
9.1k
Side Projects
sachag
452
42k
The Art of Delivering Value - GDevCon NA Keynote
reverentgeek
8
1.2k
Transcript
Scaling Sergey Gavruk @gavruk
Scaling • Ver2cal • Horizontal • By op2miza2on – Op2mize
your queries, schema, indexes – Tune you file system – Choose right disks
Share nothing architecture • Michael Stonebraker First
implementa2on in 1983 Google calls this “Sharding”
Sharding goals • App doesn’t know about clusters • Cluster
should always be available for reads and writes • Cluster should grow easily
Sharding features • Range-‐based data par22oning • Automa2c
data volume distribu2on • Transparent query rou2ng
[“a”, “g”) [“g”, “m”) [“m”, “s”) [“s”,
“z”)
[“a”, “g”) [“g”, “m”) [“m”, “s”) [“s”,
“z”) [“d”, “g”) 100 GB 500 GB 100 GB 100 GB 100 GB 400 GB 200 GB 100 GB
[“a”, “g”) [“g”, “m”) [“m”, “s”) [“s”,
“z”)
[“a”, “d”) 300 [“g”, “k”) 300 [“m”,
“s”) [“s”, “z”) 400 GB 400 GB 100 GB 100 GB [“d”, “g”) 100 [“k”, “m”) 100
None
Chunks -‐∞ +∞
Chunks -‐∞ +∞
null Numbers Strings Objects Arrays
binary data ObjectIds booleans Dates regular expressions smaller bigger
Balancing mongos balancer Config server Config
server Config server Shard 1 Shard 2
Balancer goals • keep data distributed • minimize the amount
of data transfered
Balancing mongos balancer Config server Config
server Config server Shard 1 Shard 2 Move chunk X to shard 2
Balancing Number of chunks Migra:on threshold <
20 2 21-‐80 4 80+ 8
Balancing schedule db.seangs.update({ _id : "balancer" },
{ $set : { ac2veWindow : { start : "23:00", stop : "6:00" } } }, true )
Routed Request mongos Shard 1 Shard 2
Shard 3
mongos Shard 1 Shard 2 Shard 3
Request without shard key
Without shard key + sor2ng mongos Shard 1
Shard 2 Shard 3
Consider the shard cluster if: • Data exceeds the
storage capacity of a single node • Size of working set will soon exceed your RAM • Large amount of writes
Restric2ons • You cannot update a shard key
• You must use a shard key for a single update • Index on shard key
Ideal shard key • easily divisible. • will
distribute write opera2ons among the cluster • will make it possible for the mongos to return most query opera2ons directly from a single specific mongod instance
Choosing a shard key {
_id: "1", user_id: "2345652221", date_2me: "2012-‐10-‐04“, tweet_text: “Hello world” } Reliability
Choosing a shard key Ascending { TimeStamp:
12355232, … }
Choosing a shard key Low-‐cardinality key {
Con:nent: “Europe”, Name: “Tom”, … } Zip code?
Demo
Any questions? mailto:
[email protected]