Presentation is loading. Please wait.

Presentation is loading. Please wait.

LEO Kinesis More Kafka-like Blaine Nielsen

Similar presentations


Presentation on theme: "LEO Kinesis More Kafka-like Blaine Nielsen"— Presentation transcript:

1 LEO Kinesis More Kafka-like Blaine Nielsen blaine@leoplatform.io
Open Source Project

2 Data Pipeline Monitoring
LEO Data Pipeline Monitoring Feedback via App

3 Business Intelligence
Data Needs Hierarchy Big Data AI Real Time Product Changes Predictive Analytics Deep Learning Custom Data Apps Data Science Search Customer BI Recommendation Engine Data Lake CEO BI Business Intelligence Integrations Data Warehousing Data Visualization Commission Fraud App Dashboards Evolve Simple Machine Learning Start at bottom - data in different systems, for good reason, each system has a reason and purchase it was bought. But creates the issue of keeping track of it all. Triggers between systems. BI - make sense of it all, identify trends, drive biz results Data Science - harder questions at the data Custom Data Apps - Sell custom software for more if smarter Predictive - not only know trends, but predict the future AI - real-time decision making Aggregate Understanding Explore Collect/Clean/Store

4 AWS Data Tools AI Machine Learning Custom Apps Data Science
Rekognition Lex Polly Machine Learning Machine Learning Cloud providers are giving us tools to address each need on the hierarchy. How fast can we adopt? How robust is our data management Custom Apps Lambda Data Science Athena Business Intelligence

5 Data Pipelines Legacy System Data Pipeline Data Pipeline Data Pipeline
Machine Learning Data Pipeline Using the new tools requires that we feed it the data the service needs. The services are only as good as we feed it. I see all the time. Data Scientists waiting on the org to provide access to the data. Data Pipelines take time, require maintenance. Anyone that has every worked with data knows, data changes or outside agreed upon formats. It breaks, stops flowing. So monitor to address quickly. Legacy System Data Pipeline Data Pipeline

6 Data Pipeline Spaghetti
Legacy System Machine Learning Pretty soon, we can end up with spaghetti integrations. Maintenance nightmares. Buffering, handling failure, back pressure, scaling, etc all get pushed into pipeline responsibilities

7 Reusable Data Pipeline
What if? Reusable Data Pipeline

8 Lets Try Event Driven

9 Event-driven pattern PaymentCompleted Events Payments Service
Listen and React Do and publish Payments Service OrderPlaced Events Listen & React to Events

10 “Shared Event Stream” Legacy System OrderCreated OrderCreated
Machine Learning OrderCreated Legacy System Event Stream OrderCreated OrderCreated Faster Adoption Much easier maintenance OrderCreated

11 Shared Event Stream? X X X X X X X RabbitMQ
Pub/Sub # of Subscribers Replay Durable In-Cloud Serverless Visibility X RabbitMQ unlimited X X X RabbitMQ - messaging is not a great fit Kafka - Wonderful, but really hard to get-up and running and maintenance is a pain Kinesis - in-cloud serverless is fantastic, but missing some core features Kafka unlimited Forever X X Kinesis 2 X 7 days

12 (# of Subscribers, where in log)
EventStream Kafka AWS Cloud Communication Stream Kinesis Service Catalog (# of Subscribers, where in log) DynamoDB Notification & Broker Lambda Data Store S3, DynamoDB RabbitMQ - messaging is not a great fit Kafka - Wonderful, but really hard to get-up and running and maintenance is a pain Kinesis - in-cloud serverless is fantastic, but missing some core features

13 Introducing LEO Streams
Apache2.0 OpenSource License Dev Friendly docs, SDKs, APIs, Connectors Pub/Sub unlimited subscribers Robust bursting, backoff, gzip, checkpointing, back pressure, auto retry Replay rehydrate as needed, A/B Testing AWS Bill AWS economics best practice built in Durable Store Replay Monitoring Pub/Sub LEO Streams

14 EventStream w/ AWS Producer Stream Data Store Service Catalog Consumer
200 ms 1 min Machine Learning Trigger Notify SDK read from Data Store SDK write mass Tech Map apps….. archive Data Lake Athena Compatible Choose speed w/ Auto detection and mass path based on data flow Redundancy Durable Unlimited Subscribers Bursting Backoff/retry Write Stream KAFKA-LIKE (LEO Additions) Data Store Service Catalog Broker logic Checkpoint management Tigger Broker notification Read retry Low Cost Unlimited topics on 1 Kinesis Stream

15 Stream Adoption In-App Search Mobile App LEO Streams Search UIs Old
New - 100x improvement 2M avg lag 300 MS avg lag V1 Custom APIs Legacy System In-App Search Loading Scripts LEO Streams Pub/Sub Durable Store Replay Monitoring

16 Stream Adoption In-App Search Mobile App LEO Streams Search UIs
New - 100x improvement Old 300 MS avg lag 2M avg lag V1 MicroService Custom APIs MicroService Legacy System Change Log In-App Reporting In-App Search Loading Scripts LEO Streams Pub/Sub Durable Store Replay Monitoring

17 Stream Communication LEO Streams eCommerce Search
Legacy System MicroService eCommerce Search how fast for the next micro services and SaaS system LEO Streams Pub/Sub Durable Store Replay Monitoring Legacy System MicroService Commission Engine History Log File MicroService MicroService In-App Reporting Data Warehousing

18 Monitoring UI Load Enrich offload streamline publishing
Streams Enrich Streams offload Streams streamline publishing durable store of data events, forever subscribe with ease

19 Exception Based Dashboard

20 Event Details Details Wrapper for display and audit purposes

21 to new or existing subscribers
Replay to new or existing subscribers

22 Trace Events

23 Alarms & Alerts

24 https://github.com/LeoPlatform/Nodejs
SDKs Stream Events


Download ppt "LEO Kinesis More Kafka-like Blaine Nielsen"

Similar presentations


Ads by Google