November 2023 Summaries

New AI-related announcements from re:Invent 2023 dominate the event, including new models on Bedrock, batch inference support, and updates to SageMaker, OpenSearch, Neptune, and AWS Marketplace. The newly released Bedrock foundational models include Anthropic Claude 2.1, Amazon Titan Text models, and Stable Diffusion XL 1.0, arriving at a time when the recently discovered vulnerability in ChatGPT has sparked interest. Additionally, SageMaker's HyperPod promises to reduce training time by 40% for large language model (LLM) models, while vector search is now available on various AWS databases, including DynamoDB, which can be achieved through a no-code integration with OpenSearch. Furthermore, OpenSearch has received an optimized instance family called OR1, and Neptune now supports graph analytics on existing data in Neptune or S3.

The Amazon Q chatbot is a new AI-powered tool announced by AWS, which aims to provide more up-to-date answers than ChatGPT. It's trained on AWS documentation and can be used for enterprise users to create specialized chatbots for their applications. Additionally, Q can also be used for coding, including Java developers, with features like code transformation and implementation of entire features using CodeCatalyst. Another feature is Bedrock, a managed Retrieval Augmented Generation (RAG) system that builds customized chatbots by ingesting data into a vector database. Bedrock Knowledge Base is generally available, while Bedrock Agents was announced in July 2023, enabling the creation of autonomous agents for AI applications. S3 has introduced Express One Zone storage class, offering single-digit millisecond latency and improved durability at a higher cost. Finally, AWS has also announced no-code integrations for DynamoDB and OpenSearch, although some features are limited or missing.

AWS has made several announcements, including a serverless caching solution that charges for data storage and processing units, with a minimum cost of $90 per month. Additionally, Amazon Aurora clusters can now scale to millions of write transactions per second and manage petabytes of data. SQS FIFO throughput has increased to 70,000 messages per second, and the SQS FIFO dead-letter queue (DLQ) now supports redrive via the StartMessageMoveTask API. Step Functions integrates with App Composer, allowing users to edit state machines in a workflow studio. The Go Build award for this year was given to Luc van Donkersgoed for his work on an unofficial re:Invent scheduler and the AWS News site.

AWS has made several significant announcements before re:Invent 2023, including updates to Lambda that allow it to scale 12x faster and independently, enabling the use of Lambda in situations where sudden spikes of traffic occur. Additionally, Step Functions now has public HTTP endpoints, allowing it to call any public APIs without needing a proxy or integration with API Gateway. This update also enables testing individual states programmatically without executing the state machine. Furthermore, Step Functions now integrates optimally with Bedrock, and CloudWatch introduces an AI assistant that allows users to query logs and metrics in English. Other updates include the addition of Infrequent Access log class for CloudWatch Logs, anomaly detection for CloudWatch Logs, partner integration with Adobe and Stripe for EventBridge, and synchronization between CloudFormation and Git.

The AWS re:Invent conference is approaching, bringing exciting announcements for serverless computing. CloudFormation imports have become easier, with improved support for importing resources into CloudFormation. Lambda now supports JSON logging and custom log groups, allowing for more flexibility in log management. Step Functions can restart from failure, making it a great feature for workflows that require resilience. Lambda's burst concurrency limit has been raised, enabling developers to scale their applications more easily. New runtimes have been added for Node.js 20.x and Java 21, expanding the range of programming languages supported by Lambda. CloudFront has introduced a key-value store for CloudFront Functions, allowing for faster access to static configuration data. SNS FIFO topics now support archive capabilities, enabling developers to replay messages within a specified period. Kinesis supports cross-account access, making it easier to process events with Lambda functions in other accounts. CodeWhisperer, an AI-powered code completion tool, is coming to the command line for MacOS users. EventBridge Pipes adds logging support, while EventBridge has been updated with 22 new metrics and wildcard filters. Faster scaling for SQS and Kafka has been achieved, but developers must be mindful of downstream systems. SQS now supports the JSON protocol, and CodeBuild has been updated to support Lambda. Finally, Lambda supports IPv6 for outbound connections in VPC, potentially related to upcoming IPv4 charges. These announcements bring significant improvements to serverless computing, making it easier to build, deploy, and manage scalable applications.

OpenTelemetry is an open source observability framework that offers powerful capabilities in monitoring metrics, logs, and traces. It has garnered significant attention due to its vendor-neutral approach, allowing users to instrument once and send data to any backend for processing and storage. Lumigo takes OpenTelemetry's capabilities a step further by providing near-instantaneous insights into application performance, making it easier to grasp the nuances of microservice performance. Instrumentation is essential in adding extra code to microservices to generate trace data, with manual instrumentation currently being used for languages like Golang before fully automated, 1-click OpenTelemetry becomes available. Lumigo provides a simple initialization function to set up OTel tracing, which can be easily integrated into existing services using middleware or by passing the token as an environment variable. The framework also offers visualization tools, allowing users to gain deep insights into service performance without the burden of running their own trace storage and processing. OpenTelemetry is invaluable for tracking down performance issues or troubleshooting unexpected behavior in Golang microservices, offering flexibility and operational overhead benefits.

The KubeCon North America conference is a pivotal event for cloud-native communities, showcasing the latest innovations and advancements in open-source philosophy. The conference featured inspiring keynote sessions, including those by Priyanka Sharma and Taylor Dolezal, which painted vivid pictures of the cloud-native landscape present and future. Sessions on environmental sustainability in the cloud, CNCF Graduated Project Updates, and Kubernetes Project Updates provided valuable insights into the latest developments within the CNCF landscape. The Observability Technical Advisory Group session was a hub for collaborative discussion focused on observability of cloud-native ecosystems, with updates on pivotal projects like observe-k8s and query language standardization. The conference also highlighted the importance of community spirit, encouraging attendees to get involved in contributing to documentation, code development, or participating in future events. Overall, KubeCon North America 2023 was a remarkable event that underscored the advancement and togetherness of cloud-native communities.

As a developer working with Kubernetes, understanding the metrics that illustrate what's happening behind the scenes is vital for a smooth-sailing cluster. Monitoring key metrics such as backoff errors and their relation to pod crash loops, container errors, node resource utilization, API server errors, and DaemonSet misschedules can help identify potential issues before they escalate. By using tools like Lumigo, developers can receive real-time alerts for critical events, pinpoint the root cause of problems, and take corrective actions to optimize application performance and ensure system reliability.