Working on a lot of bleeding edge technologies • Passionate developer: love writing code, hate meetings The Neosperience Cloud • Software as a service cloud for Digital Customer Experience processes (Psychographics, Loyalty & Gamification, Proximity, Content, etc.) • Built on AWS, 95% on Serverless technologies • Moved from VMWare, to EC2, to Elastic Beanstalk, to Serverless • Dozens of micro and nano services The Digital Customer Experience Company, aims to change the way brands and customers interact with an approach of a software vendor targeting Digital Customer Experience as the evolution of Marketing automation.
ephemeral compute power that comes into existence on request and disappears immediately after use. Use of this architecture can mitigate some security concerns such as security patching and SSH access control, and can make much more efficient use of compute resources. These systems cost very little to operate and can have inbuilt scaling features.” — ThoughtWorks, 2016
great way to decouple services • largely abused in the last decade (ESB) • now revamped with CQRS and Event Sourcing to handle microservices communication • serverless functions handle events • need a different approach on architectures
set of FB auth tokens are sent to our system 2. jobs to download user images and posts are started 3. downloaded images are classified through Rekognition 4. posts are processed to perform text mining 5. downloaded profile is updated 6. completion acknowledge sent back to the caller (with stats) many architectures choices, how to evaluate them..?
Compute, presented to the ServerlessConf the so-called Serverless Manifesto. • In 2017 Danilo Poccia, Technical Evangelist @ AWS, talking at JeffConf asked “what would be having that manifesto translated into a checklist to evaluate Serverless compliance of an architecture?” • So, we turned it into a “Serverless Scorecard” Evaluating a serverless architecture (aka how big is my serverless?) The Serverless Scorecard Function as the unit of deployment and scaling Implicitly fault-tolerant Metrics No machines, VMs, or containers Bring Your Own Code Stateless Never pay for idle Scales per request
distributed monolith with a single endpoint receiving a set of FB tokens and mapping to a Lambda function invoking a workflow for each user (made of many lambdas) Pains: • Orchestrator lambda has to wait until the last workflow has been completed • NodeJS works better than other languages due to its asynchronous programming model IDEA 01 Orchestrator Job Orchestrator DownloadPosts DownloadImages ProcessImages Job Orchestrator DownloadPosts DownloadImages ProcessImages [ … one for each user … ]
on FB token. Lambda are invoked asynchronously using DynamoDB streams. Workflow completion is done as an update on Table record Pains: • Orchestrator lambda has to wait until the last workflow has been completed (checked on Dynamo) • Each Lambda still has to wait for other lambda to have finished IDEA 02 Orchestrator Job Orchestrator DownloadPosts DownloadImages ProcessImages
and DynamoDB Table to check completion of a task for a specific user. Dynamo and Kinesis keep the state of the system Pains: • Orchestrator lambda has to wait until the last workflow has been completed (checked on Dynamo) • Can’t know at which my system current state IDEA 03 Orchestrator DownloadPosts DownloadImages ProcessImages
trigger different Lambdas, defer to Kinesis only batch execution within a single task. Implement End-of-Task event within Kinesis to signal processing is completed Pains: • Orchestrator lambda has to wait until the last workflow has been completed (checked on Dynamo) • Kinesis parallelism is through shard and partition key (i.e. userId) IDEA 04 Orchestrator DownloadPosts DownloadImages ProcessImages
Write a file for each computation step and let streams to parallel trigger Lambda invocation. Use SepFunction to check processing status globally (configure polling through wait/check states) Pains: • Client needs to use AWS SDK and conform to S3 data model or a Lambda has to proxy it • No event failover if one file is lost • No event replay from Event Store IDEA 05 Write file with token SplitPostsJobs DownloadPosts DownloadPosts DownloadPosts [ … ] Check/Wait files on bucket DownloadImage(s) DownloadImage(s) DownloadImage(s) [ … ] ProcessImage(s) ProcessImage(s) ProcessImage(s) SplitImagesJob(s) WriteEndFile
SDK and conform to S3 data model or a Lambda has to proxy it • No event failover if one file is lost • No event replay from Event Store Is a better solution possible?
Lambda to be notified when a new event is available (either a job request or a response) • IoT topics can be created on-the-fly so no CF setup is needed, just having IoT active in AWS account • Functions can subscribe to any topic through IoT Rules any Lambda can be triggered by a new event • Firehose can persist events into S3 event replay from EventStore (S3) and failover can be achieved (and it’s Serverless!) • Async implicitly supported • REST endpoints decoupling A Lambda connected to APIG generates a requestID and returns from HTTP request with a token to poll for completion
Implicitly fault-tolerant (improved by allowing event replay from Event Store) Metrics No machines, VMs, or containers Bring Your Own Code Stateless Never pay for idle Scales per request REST compatible 9/8
aletheia
shinrinakamura
lycorptech_jp
puhitaku
maimyyym
phaya72
matsuihidetoshi
akagire
itkq
toru_kubota
macloud
yushin_n
haochenx
marcduiker
denniskardys
palkan
jrom
bermonpainter
searls
soudai
notwaldorf
mongodb
bkeepers
marktimemedia
akmur
