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
AWS SQS queues & Kubernetes Autoscaling Pitfall...
Search
Eric Khun
October 26, 2020
Programming
2
450
AWS SQS queues & Kubernetes Autoscaling Pitfalls Stories
Talk at the Cloud Native Computing Foundation meetup @dcard.tw
Eric Khun
October 26, 2020
Tweet
Share
More Decks by Eric Khun
See All by Eric Khun
From PHP to Golang: Migrating a real-time data replication service
erickhun
1
130
Other Decks in Programming
See All in Programming
Protocol Buffersの型を超えて拡張性を得る / Beyond Protocol Buffers Types Achieving Extensibility
linyows
0
100
UbieのAIパートナーを支えるコンテキストエンジニアリング実践
syucream
2
800
もうちょっといいRubyプロファイラを作りたい (2025)
osyoyu
0
230
開発チーム・開発組織の設計改善スキルの向上
masuda220
PRO
18
9.7k
パッケージ設計の黒魔術/Kyoto.go#63
lufia
3
410
Microsoft Orleans, Daprのアクターモデルを使い効率的に開発、デプロイを行うためのSekibanの試行錯誤 / Sekiban: Exploring Efficient Development and Deployment with Microsoft Orleans and Dapr Actor Models
tomohisa
0
230
フロントエンドのmonorepo化と責務分離のリアーキテクト
kajitack
2
160
レガシープロジェクトで最大限AIの恩恵を受けられるようClaude Codeを利用する
tk1351
4
1.6k
Improving my own Ruby thereafter
sisshiki1969
1
150
プロポーザル駆動学習 / Proposal-Driven Learning
mackey0225
1
330
【第4回】関東Kaggler会「Kaggleは執筆に役立つ」
mipypf
0
1k
Rancher と Terraform
fufuhu
2
200
Featured
See All Featured
We Have a Design System, Now What?
morganepeng
53
7.8k
Agile that works and the tools we love
rasmusluckow
330
21k
A Modern Web Designer's Workflow
chriscoyier
696
190k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
36
2.5k
The Art of Programming - Codeland 2020
erikaheidi
55
13k
How to Think Like a Performance Engineer
csswizardry
26
1.9k
Speed Design
sergeychernyshev
32
1.1k
Rails Girls Zürich Keynote
gr2m
95
14k
The Cult of Friendly URLs
andyhume
79
6.6k
Rebuilding a faster, lazier Slack
samanthasiow
83
9.1k
A Tale of Four Properties
chriscoyier
160
23k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
48
9.7k
Transcript
AWS SQS queues & Kubernetes Autoscaling Pitfalls Stories Cloud Native
Foundation meetup @dcard.tw @eric_khun
Make it work, Make it right, Make it fast kent
beck (agile manifesto - extreme programming)
Make it work, Make it right, Make it fast kent
beck (agile manifesto - extreme programming)
Make it work, Make it right, Make it fast kent
beck (agile manifesto - extreme programming)
Buffer
None
Buffer • 80 employees , 12 time zones, all remote
Quick intro
None
Main pipelines flow
it can look like ... golang Talk @Maicoin :
None
None
How do we send posts to social medias?
A bit of history... 2010 -> 2012: Joel (founder/ceo) 1
cronjob on a Linode server $20/mo 512 mb of RAM 2012 -> 2017 : Sunil (ex-CTO) Crons running on AWS ElasticBeanstalk / supervisord 2017 -> now: Kubernetes / CronJob controller
AWS Elastic Beanstalk: Kubernetes:
At what scale? ~ 3 million SQS messages per hour
Different patterns for many queues
Are our workers (consumers of the SQS queues ) efficients?
Are our workers efficients?
Are our workers efficients?
Empty messages? > Workers tries to pull messages from SQS,
but receive “nothing” to process
Number of empty messages per queue
Sum of empty messages on all queues
None
1,000,000 API calls to AWS costs 0.40$ We have 7,2B
calls/month for “empty messages” It costs ~$25k/year > Me:
None
AWS SQS Doc
None
Or in the AWS console
Results?
empty messages
AWS
None
$120 > $50 saved daily > $2000 / month >
$25,000 / year (it’s USD, not TWD)
Paid for querying “nothing”
(for the past 8 years )
Benefits - Saving money - Less CPU usage (less empty
requests) - Less throttling (misleading) - Less containers > Better resources allocation: memory/cpu request
Why did that happen?
Default options
None
Never questioning what’s working decently or the way it’s been
always done
What could have helped? Infra as code (explicit options /
standardization) SLI/SLOs (keep re-evaluating what’s important) AWS architecture reviews (taging/recommendations from aws solutions architects)
Make it work, Make it right, Make it fast
Make it work, Make it right, Make it fast
Do you remember?
None
None
None
Need to analytics on Twitter/FB/IG/LKD… on millions on posts faster
workers consuming time
None
What’s the problem?
Resources allocated and not doing anything most of the time
Developer trying to put find compromises on the number of workers
How to solve it?
Autoscaling! (with Keda.sh) Supported by IBM / Redhat / Microsoft
None
Results
None
But notice anything?
Before autoscaling
After autoscaling
After autoscaling
What’s happening?
Downscaling
Why?
delete pod lifecycle
what went wrong - Workers didn’t manage SIGTERM sent by
k8s - Kept processing messages - Messages were halfway processed and killed - Messages were sent back to the the queue again - Less workers because of downscaling
solution - When receiving SIGTERM stop processing new messages -
Set a graceful period long enough to process the current message if (SIGTERM) { // finish current processing and stop receiving new messages }
None
None
And it can also help with sqs empty messages
Make it work, Make it right, Make it fast
Make it work, Make it right, Make it fast
Thanks!
Questions? monitory.io taiwangoldcard.com travelhustlers.co ✈