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
etcd - mission-critical key-value store - CoreO...
Search
Brandon Philips
May 09, 2016
Programming
3
370
etcd - mission-critical key-value store - CoreOS Fest 2016
Brandon Philips
May 09, 2016
Tweet
Share
More Decks by Brandon Philips
See All by Brandon Philips
Node.js Workflow with Minikube and Skaffold
philips
0
220
Manage the App on Kubernetes
philips
0
310
Production Backbone Monitoring Containerized Apps
philips
0
120
KubeCon EU 2017: Dancing on the Edge of a Volcano
philips
1
620
rkt - KubeCon EU keynote - 2017
philips
1
240
FOSDEM_Keynote_2017-_.pdf
philips
0
90
Tectonic Summit Day 2 Keynote
philips
0
310
Kubernetes: Simple to Manage Anywhere (self-hosted, Tectonic upgrade demo)
philips
0
280
KubeCon Keynote 2016- Distributed Systems Simplified on Kubernetes
philips
2
530
Other Decks in Programming
See All in Programming
WebフロントエンドにおけるGraphQL(あるいはバックエンドのAPI)との向き合い方 / #241106_plk_frontend
izumin5210
4
1.4k
Why Jakarta EE Matters to Spring - and Vice Versa
ivargrimstad
0
1.1k
Enabling DevOps and Team Topologies Through Architecture: Architecting for Fast Flow
cer
PRO
0
340
OSSで起業してもうすぐ10年 / Open Source Conference 2024 Shimane
furukawayasuto
0
110
PHP でアセンブリ言語のように書く技術
memory1994
PRO
1
170
Kaigi on Rails 2024 〜運営の裏側〜
krpk1900
1
230
Jakarta EE meets AI
ivargrimstad
0
590
Webの技術スタックで マルチプラットフォームアプリ開発を可能にするElixirDesktopの紹介
thehaigo
2
1k
광고 소재 심사 과정에 AI를 도입하여 광고 서비스 생산성 향상시키기
kakao
PRO
0
170
とにかくAWS GameDay!AWSは世界の共通言語! / Anyway, AWS GameDay! AWS is the world's lingua franca!
seike460
PRO
1
890
Hotwire or React? ~アフタートーク・本編に含めなかった話~ / Hotwire or React? after talk
harunatsujita
1
120
初めてDefinitelyTypedにPRを出した話
syumai
0
420
Featured
See All Featured
Product Roadmaps are Hard
iamctodd
PRO
49
11k
Designing for Performance
lara
604
68k
Bootstrapping a Software Product
garrettdimon
PRO
305
110k
Java REST API Framework Comparison - PWX 2021
mraible
PRO
28
8.2k
Building an army of robots
kneath
302
43k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
33
1.9k
How To Stay Up To Date on Web Technology
chriscoyier
788
250k
The Cost Of JavaScript in 2023
addyosmani
45
6.8k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
42
9.2k
4 Signs Your Business is Dying
shpigford
180
21k
5 minutes of I Can Smell Your CMS
philhawksworth
202
19k
Rebuilding a faster, lazier Slack
samanthasiow
79
8.7k
Transcript
Brandon Philips @BrandonPhilips |
[email protected]
etcd - mission-critical key-value store
Uncoordinated Upgrades
... ... ... ... ... ... Unavailable Uncoordinated Upgrades
Motivation CoreOS cluster reboot lock - Decrement a semaphore key
atomically - Reboot and wait... - After reboot increment the semaphore key
3 CoreOS updates coordination
CoreOS updates coordination 3
... CoreOS updates coordination 2
... ... ... CoreOS updates coordination 0
... ... ... CoreOS updates coordination 0
... ... CoreOS updates coordination 0
... ... CoreOS updates coordination 0
... ... CoreOS updates coordination 0
... ... ... CoreOS updates coordination 0
CoreOS updates coordination
Store Application Configuration config
config Start / Restart Start / Restart Store Application Configuration
config Update Store Application Configuration
config Unavailable Store Application Configuration
Requirements Strong Consistency - mutual exclusive at any time for
locking purpose Highly Available - resilient to single points of failure & network partitions Watchable - push configuration updates to application
Requirements CAP - We want CP - We want something
like Paxos
Common problem GFS Paxos Big Table Spanner CFS Chubby Google
- “All” infrastructure relies on Paxos
Common problem Amazon - Replicated log powers ec2 Microsoft -
Boxwood powers storage infrastructure Hadoop - ZooKeeper is the heart of the ecosystem
COMMON PROBLEM #GIFEE and Cloud Native Solution
10,000 Stars on Github 250 contributors Google, Red Hat, EMC,
Cisco, Huawei, Baidu, Alibaba...
THE HEART OF CLOUD NATIVE Kubernetes, Cloud Foundry Diego, Project
Calico, many others
ETCD KEY VALUE STORE Fully Replicated, Highly Available, Consistent
PUT(foo, bar), GET(foo), DELETE(foo) Watch(foo) CAS(foo, bar, bar1) Key-value Operations
DEMO play.etcd.io
Runtime Reconfiguration Point-in-time Backup Extensive Metrics etcd Operationality
ETCD v3 Successor of etcd v2
ETCD v3 Better Performance
ETCD v3 More Efficient APIs
Multi-Version Put(foo, bar) Put(foo, bar1) Put(foo, bar2) Get(foo) -> bar2
Multi-Version Put(foo, bar) Put(foo, bar1) Put(foo, bar2) Get(foo, 1) ->
bar
Tx.If( Compare(Value("foo"), ">", "bar"), Compare(Version("foo"), "=", 2), ... ).Then( Put("ok","true")...
).Else( Put("ok","false")... ).Commit() Mini-Transactions
l = CreateLease(15 * second) Put(foo, bar, l) l.KeepAlive() l.Revoke()
Leases
w = Watch(foo) for { r = w.Recv() print(r.Event) //
PUT print(r.KV) // foo,bar } Streaming Watch
Synchronization LoC
ETCD v2 machine coordination -> O(10k)
ETCD v3 app/container coordination -> O(1M)
Performance 1K keys
Performance Snapshot caused performance degradation etcd2 - 600K keys
Performance etcd2 - 600K keys Snapshot triggered elections
ZooKeeper Performance Non-blocking full snapshot Efficient memory management
Performance ZooKeeper default
Performance Snapshot triggered election ZooKeeper default
Performance Snapshot ZooKeeper default
Performance GC ZooKeeper snapshot disabled
Reliable Performance - Similar to ZooKeeper with snapshot disabled -
Incremental snapshot - No Garbage Collection Pauses - Off-heap storage
Performance etcd3 /ZooKeeper snapshot disabled
Performance etcd3 /ZooKeeper snapshot disabled
Memory 10GB 2.4GB 0.8GB 512MB data - 2M 256B keys
Reliability 99% at small scale is easy - Failure is
infrequent and human manageable 99% at large scale is not enough - Not manageable by humans 99.99% at large scale - Reliable systems at bottom layer
HOW DO WE ACHIEVE RELIABILITY WAL, Snapshots, Testing
Write Ahead Log Append only - Simple is good Rolling
CRC protected - Storage & OSes can be unreliable
Snapshots Torturing DBs for Fun and Profit (OSDI2014) - The
simpler database is safer - LMDB was the winner Boltdb an append only B+Tree - A simpler LMDB written in Go
Testing Clusters Failure Inject failures into running clusters White box
runtime checking - Hash state of the system - Progress of the system
Testing Cluster Health with Failures Issue lock operations across cluster
Ensure the correctness of client library
TESTING CLUSTER dash.etcd.io
etcd/raft Reliability Designed for testability and flexibility Used by large
scale db systems and others - Cockroachdb, TiKV, Dgraph
etcd vs others Do one thing
etcd vs others Only do the One Thing
etcd vs others Do it Really Well
etcd Reliability Do it Really Well
ETCD v3.0 BETA Efficient and Scalable
BETA AVAILABLE TODAY github.com/coreos/etcd
FUTURE WORK Proxy, Caching, Watch Coalescing, Secondary Index
GET INVOLVED github.com/coreos/etcd
Brandon Philips @BrandonPhilips |
[email protected]
etcd - mission-critical key-value store
Thank you!