Hi, here’s a quick round-up of my November.
re:Invent 2023 recap
Here is the best serverless news from re:Invent 2023.
Each Lambda function can now burst to 1000 concurrent executions every 10 seconds, independently. It means it will take a single function just a minute to reach a peak concurrency of 6,000 concurrent executions. Every function can burst independently. So assuming your account level concurrency limit allows, multiple functions can burst to the 6,000 concurrent executions within that same minute. This shatters the previous account-wide burst concurrency limit of 3000, followed by an increment of 500 per minute.
This means Lambda is now suitable for use cases that experience very bursty traffic! But if the high level of throughput is sustained (rather than in a burst) then you should still consider moving the workload to containers for better cost efficiency.
You can now call 3rd party APIs directly from your state machine.
You can use this new endpoint to test individual states without deploying the whole state machine. This is especially useful for those hard-to-reach branches of state machines. I’m excited about this feature and will update my Testing Serverless Architectures course shortly.
You can now use English to query CloudWatch logs and metrics. In my testing, it often doesn’t get the query right the first time. For example, in the example below, the generated query returns the function with the most invocations, not cold starts.
I wouldn’t trust this AI assistant’s output at face value, but it can still give you a good starting point and save you some typing.
If you don’t need any of these features, then the new infrequent access log class can save you 50% on log ingestion costs. It’s a shame you can’t use subscription filters. This can impact many vendors who use subscription filters today to process and forward your logs.
Ok, not a serverless news per se, as this is only applicable to EC2, ECS and EKS and only works for Java applications. But it’s a good first step from AWS to provide automated instrumentation for customer applications.
That is, you can now redrive messages from an SQS FIFO DLQ to the source SQS FIFO queue for reprocessing.
AWS’s new chat assistant. Available in AWS documentation, AWS console as well as your IDE (via the AWS Toolkit plugin).
It can provide contextual answers – if you talk to it in the AWS console, it will have access to your AWS resources, and if you talk to it in VS Code then it can help you with your code.
From my testing thus far, its answers are hit-and-miss. Subjectively, it feels less accurate compared to ChatGPT and it hallucinates more.
But it’s free.
Bedrock supports new foundational models
- Anthropic Claude 2.1
- Amazon Titan Text models (Lite & Express)
- Amazon Titan Multimodal Embeddings model
- Amazon Titan Image Generator model
- Stable Diffusion XL 1.0
- Meta Llama 2 70B
And promises to reduce training time by 40%!
But remember, OpenSearch Serverless has a minimum charge of $700/month!
This is a nice update. You don’t have to write custom code to sync DynamoDB changes to OpenSearch anymore.
It promises to give you a 30% better price-to-performance ratio.
“enables you to scale your Amazon Aurora clusters to millions of write transactions per second and manage petabytes of data”
Not as bad as OpenSearch “serverless”, but this one comes with a minimum cost of $90/month even if you have no data.
If you really need a serverless cache, go check out Momento instead.
This is not for everyone, but there are lots of small admin/ops/security stacks that can benefit from this.
Data is still replicated across multiple disks and retains the 11 nines durability. But you can suffer data loss if the data centre is destroyed. I’m surprised the calculation of durability doesn’t include some probability for such (highly unlikely) scenarios.
It promises single-digit ms latency. The cost per million requests is 50% cheaper, but the storage cost is 7x the cost of the standard class.
This is not for everyone, unless you need low latency (single digit ms) access to relatively small amounts of data.
If you work with AWS SAM or CloudFormation templates, then you can use Application Composer to visualize and design your template directly in VS Code (via the AWS Toolkit plugin).
I published seven blog posts.
- How I implemented web analytics reporting with Timestream
- The best pre:invent serverless announcements you may have missed
- re:Invent 2023 day 0 recap
- re:Invent 2023 day 1 recap
- re:Invent 2023 day 2 recap
- re:Invent 2023 day 3 recap
- re:Invent 2023 day 4 recap
We aired five episodes on the Real-World Serverless podcast:
- #87 – Terraform expert on Serverless.TF and the Terraform licensing fiasco [Anton Babenko]
- #88 – Improving Lambda cold starts with AJ Stuyvenberg
- #89 – Hashnode’s Scalable Serverless Architecture [Sandro Volpicella]
- #90 – He built a hotel booking system that costs $0.82/month to run [Hieu Do]
- #91 – Father of DevOps on the future of AI and DevOps [Patrick DeBois]
I published eleven videos.
- The Most Costly Serverless Mistake is…
- How to implement caching for serverless applications
- API Gateway vs Application Load Balancer (ALB) cost comparison
- How I built an affiliate tracking system in a weekend with serverless
- AWS CDK: the value of snapshot testing
- AWS Lambda: monolithic vs single-purpose functions
- How CyberArk test serverless architectures at scale
- Charity Majors on DevOps culture and team dynamic
- How CyberArk achieves rock solid Tenant Isolation
- The hardest part of CDK, according to an AWS DevTools Hero
- AWS CDK: problem with creating reusable L3 constructs
If you like the topics I cover then please subscribe to my channel and never miss a thing.
I had a chat with Waldemar Hummer, CTO of LocalStack, on some key takeaways from running and debugging serverless applications locally. I have been sceptical of LocalStack for a long time, but it’s looking slick these days. It even simulates IAM checks and can identify permission-related problems.
The webinar was on the 9th of November, and the recording should be available soon.
Looking ahead to December
Ok, November was crazy! I’m taking some time off in December to catch up on re:Invent sessions and update my courses. If you’re taking part in the January cohort of Production-Ready Serverless, then expect plenty of brand-new lectures and all the exercises will be available in CDK as well as the Serverless Framework.
On December 11th, I will be cohosting a free webinar with Khawaja Shams (CEO of Momento) and Aviad Mor (CTO of Lumigo). Hope to see you there!
Until then, ciao :-)