re:Invent 2024: A DevOps Engineer's Take

re:Invent 2024 wrapped up earlier this month. As always, the keynotes were dense with announcements, some that genuinely change how we build on AWS, and many that won’t.

This post is our attempt to filter the signal. Five announcements we think actually matter for the day-to-day work of running production workloads on AWS.

1. S3 Tables

Amazon S3 Tables is a managed Apache Iceberg storage service. Iceberg has been the de facto open table format for a while, but running it yourself (catalog, compaction, optimization) is real operational work. S3 Tables takes that on.

For DevOps teams supporting analytics workloads, this is a meaningful reduction in surface area. No more babysitting compaction Spark jobs. Query engines like Athena and EMR can read directly.

Worth investigating for any team running self-managed Iceberg or contemplating Hudi. Probably not worth migrating Glue tables that are already running smoothly.

2. Aurora DSQL (preview)

Aurora DSQL is a distributed, serverless, multi-region active-active SQL database, PostgreSQL-compatible. The pitch: strong consistency across regions, no primary, no failover.

It’s preview-only right now and the feature set is intentionally narrow. But this fills a real gap in the AWS database lineup. We’ve had clients building on DynamoDB Global Tables when they really wanted relational semantics; Aurora DSQL is aimed at them.

Wait for GA before betting production workloads on it. But put it on your radar.

3. Amazon Q Developer agents

The Q Developer feature set picked up agentic capabilities: agents that can perform multi-step tasks like generating unit tests across a codebase, transforming Java versions, or doing automated code reviews.

For DevOps work, the relevant question is when (not if) the same approach gets applied to infrastructure. We expect to see agents that can iteratively author and validate IaC, run cfn-lint, fix failures, and produce reviewable PRs. The pieces are there.

4. S3 Metadata

S3 Metadata gives you queryable metadata about your bucket contents (object keys, sizes, last-modified, custom user metadata) exposed as an Iceberg table. No more S3 Inventory CSVs piped through Athena to find what’s in your buckets.

This is one of those quietly useful announcements. If you’ve ever written a Lambda that paginates through ListObjectsV2 looking for files matching some criteria, S3 Metadata is going to make your life better.

5. Trainium2 and Bedrock improvements

We don’t usually cover model news in DevOps posts, but two threads worth tracking:

• Trainium2 instances are now available, with strong price/performance for inference workloads.
• Bedrock added more model families and prompt caching improvements, which materially affects cost when you’re running LLM features at scale.

If you’re building anything with LLMs in production, both of these change the cost math.

What we’re skipping

A lot of keynote-time announcements aren’t going to affect most teams in 2025: many of the SageMaker analytics integrations, the Q Business updates, much of the partner-focused tooling. Useful for some, not load-bearing for most.

What we’re doing in 2025

Internally, we’re prioritizing:

• Production trials of Bedrock-based features for client-facing internal tools
• S3 Tables migration evaluation for clients running self-managed Iceberg
• Watching Aurora DSQL toward GA

If you want to talk through how any of this affects your AWS architecture, get in touch.