Sunday, July 26, 2026

MariaDB announced that MySQL Galera Cluster reaches end of life on September 30, 2026, ceasing maintenance and binary releases for that build. New clustering features will exclusively target MariaDB Galera Cluster. Percona provides guidance for migrating to Percona XtraDB Cluster as an alternative path.

• MySQL Galera Cluster stops receiving updates after September 30, 2026.
• MariaDB directs users to migrate to MariaDB Galera Cluster for new features.
• Percona offers Percona XtraDB Cluster as a migration target for MySQL users.
• Plan migration strategy now to avoid unsupported software risks.

With UPDATE/DELETE FOR PORTION OF likely arriving in Postgres 19, Paul Jungwirth is studying Richard Snodgrass's 1993 TQuel paper to understand temporal relational algebra. The goal is to identify algebraic identities that allow the query planner to transform queries into more efficient execution shapes. This research focuses on optimizing operations like filtering rows before joins to improve performance for temporal data.

• Postgres 19 is expected to support UPDATE/DELETE FOR PORTION OF, enabling temporal row modifications.
• Jungwirth is leveraging 1993 TQuel research to derive algebraic identities for the modern query planner.
• Understanding these identities helps the planner push filters before joins for better performance.
• This work connects historical temporal database theory to current Postgres optimization strategies.

Amazon RDS now supports MySQL 9.7 in its Database Preview Environment, allowing users to test the latest Long-Term Support release before general availability. This sandbox enables evaluation of new features and security patches specific to the community edition on RDS. Instances are retained for up to 60 days and automatically deleted afterward.

• MySQL 9.7 is the newest LTS release, including bug fixes and security patches.
• Preview instances are sandboxed and auto-deleted after 60 days.
• Use this environment to validate application compatibility with MySQL 9.7.
• Snapshots created in the preview environment can be retained beyond the instance lifecycle.

The Inflect-Micro-v2 model achieves full voice generation capabilities with a parameter count of just 9.36 million. This compact architecture demonstrates that high-fidelity voice synthesis does not require massive model sizes. The release highlights a trend toward efficient, lightweight models for audio tasks.

• 9.36M parameters enable complete voice synthesis, challenging size norms.
• Ideal for edge deployment where compute and memory are constrained.
• Reduces inference costs and latency for voice-enabled applications.
• Demonstrates efficient architecture design for audio generation tasks.

The Debian Project is currently debating three distinct proposals regarding the integration of Large Language Models into their development processes. These proposals aim to establish clear guidelines on how AI tools can be utilized by contributors and maintainers. The vote represents a significant step in defining the ethical and technical boundaries of AI adoption within a major open-source distribution.

• Debian is formalizing AI tool usage through a formal voting process.
• Three specific proposals define scope, ethics, and implementation details.
• Outcome will set precedent for other major open-source projects.
• Impacts contributor workflows and maintainership guidelines.

AWS has made Claude Opus 5 available via Amazon Bedrock, featuring enhanced coding capabilities and long-running agent support. The model introduces zero data retention compliance, making it suitable for sensitive enterprise workloads. It offers improved reasoning for complex analysis and navigation of large codebases.

• Claude Opus 5 is now live on Amazon Bedrock for enterprise customers.
• Model supports zero data retention (ZDR) for strict compliance needs.
• Enhanced coding skills include adaptive strategy and production-quality output.
• Agents can run for hours with error recovery and obstacle navigation.
• Significant accuracy gains for long-document and complex analysis tasks.

Expedia Group has deployed STAR, an internal platform leveraging LLMs to accelerate production incident investigations. The system integrates service telemetry with structured workflows to generate root cause assessments and guide engineers through response protocols. Built on a stack including FastAPI, Datadog, and Langfuse, it maintains human oversight while automating initial analysis steps.

• STAR uses LLMs to analyze telemetry and propose root causes for production incidents.
• The tool keeps engineers in the loop, supporting rather than replacing human decision-making.
• Architecture relies on FastAPI, Datadog, Celery, Redis, and Langfuse for observability and orchestration.
• Structured workflows ensure consistent analysis and faster incident resolution times.

aisuite provides a lightweight Python library to simplify interactions with various generative AI providers. It supports bringing your own API keys for services like OpenAI, Anthropic, and Google, or running models locally via Ollama. The project also spawned OpenWorker, a desktop AI coworker that handles tasks like file reading, communication integration, and automation while keeping data on-premises.

• Unified API abstraction for OpenAI, Anthropic, Google, and local Ollama models
• Lightweight Python library reduces boilerplate for integrating multiple LLMs
• OpenWorker desktop app leverages aisuite for local-first AI task automation
• Supports macOS 13+ (Apple Silicon) and Windows 10/11 for local deployment

This repository provides a collection of agent skills designed for real-world application development, contrasting with broader process frameworks like GSD or Spec-Kit. The author argues that such frameworks often remove developer control and complicate bug resolution. These skills are intended to be small, adaptable, and model-agnostic, drawing on extensive engineering experience rather than generic 'vibe coding' approaches.

• Prioritizes developer control over rigid process frameworks like GSD or BMAD.
• Skills are small, composable, and designed to work with any AI model.
• Targets complex application development rather than simple code generation.
• Easy setup via npx installer for immediate integration into agent workflows.

AWS has introduced aws-bench, an open-source benchmark designed to evaluate how accurately and efficiently AI agents perform real-world infrastructure tasks. The suite includes test cases derived from actual AWS usage patterns, covering troubleshooting, investigation, and resource creation. Each case links natural language queries to specific cloud states and ground-truth answers, enabling consistent scoring for model providers and researchers.

• Use aws-bench to objectively measure AI agent accuracy and efficiency on AWS infrastructure.
• Access a public suite of test cases based on real-world AWS usage scenarios.
• Diagnose agent failures by comparing performance against defined ground-truth answers.
• Leverage reproducible benchmarks to improve foundation models for cloud operations.

Amazon Bedrock AgentCore now routes agent telemetry, including traces, prompts, and structured logs, into a single per-agent CloudWatch log group. Previously, trace spans and event logs were split across multiple destinations, complicating debugging workflows. This consolidation allows for unified observability and simplifies access control and encryption management at the individual agent level.

• Unified observability: Traces, prompts, and logs now reside in one log group per agent.
• Simplified debugging: No need to search across multiple log groups for agent invocation details.
• Granular security: Enables fine-grained access control and CMK encryption per agent.
• Streamlined ops: Reduces operational overhead by consolidating telemetry destinations.

Amazon MWAA now supports Apache Airflow 2.11.2, a maintenance release focused on security and stability. This update patches core dependencies, improves task lifecycle management for queued tasks, and enhances secrets masking in logs. It also includes UI corrections and provider package updates for S3 and CloudWatch log delivery.

• MWAA environments can now be created or upgraded to Airflow 2.11.2.
• Security patches applied to core dependencies and webserver layers.
• Fixed task lifecycle management issues for queued tasks.
• Enhanced secrets masking reduces sensitive data exposure in logs.
• Updates include S3 and CloudWatch log provider packages.

Percona has updated PBM 2.15.0 and PCSM 0.9.0 to include CycloneDX 1.6 Software Bill of Materials in all release artifacts, including binaries and Docker images. This allows teams to immediately scan for vulnerabilities using compatible tools like Trivy or Grype upon deployment. The integration streamlines security compliance by providing machine-readable dependency data from day one.

• PBM 2.15.0 and PCSM 0.9.0 include CycloneDX 1.6 SBOMs in all artifacts.
• Scan Docker images quickly with Trivy using the --sbom-sources oci flag.
• Use any CycloneDX-compatible tool like Grype to detect CVEs immediately.
• SBOMs are available in tarballs, RPMs, DEBs, and container images.

AWS has made the Durable Execution SDK for .NET generally available, allowing C# developers to build resilient, long-running workflows directly within Lambda. This SDK enables automatic progress checkpointing and execution pauses of up to a year for external events, eliminating the need for custom tracking or external orchestration services. It supports complex patterns like payment pipelines, AI agent orchestration, and human-in-the-loop approvals using an idiomatic C# interface.

• Build long-running Lambda workflows without external orchestration tools.
• Automatic checkpointing ensures resilience against unexpected failures.
• Execution can pause for up to a year while waiting on events.
• Idiomatic C# SDK simplifies development of durable functions.
• Ideal for payment pipelines, AI agents, and approval workflows.

Zalando engineered an in-process, client-side load balancer to manage a high-throughput API processing approximately one million requests per second. This architectural shift delivered more predictable latency characteristics and reduced overall infrastructure expenses. The implementation also provided superior observability, allowing the team to pinpoint failure origins with greater precision.

• Client-side load balancing shifts decision logic from network edge to application process.
• Predictable latency improves significantly under high load compared to traditional proxies.
• Infrastructure costs drop by removing dedicated load balancing hardware or services.
• Enhanced visibility allows faster root cause analysis for distributed failures.

Alibaba has open-sourced OpenCodeReview, an AI-powered CLI tool originally used internally to serve tens of thousands of developers. The tool employs a hybrid architecture combining deterministic pipelines with an LLM Agent to analyze Git diffs and identify defects. It supports OpenAI and Anthropic compatible endpoints and includes a built-in ruleset for issues like NPE, thread-safety, XSS, and SQL injection.

• Hybrid architecture merges deterministic checks with LLM agents for precise line-level comments.
• Battle-tested at Alibaba scale, serving tens of thousands of developers over two years.
• Configurable model endpoints support OpenAI and Anthropic compatible interfaces.
• Built-in fine-tuned ruleset detects NPE, thread-safety, XSS, and SQL injection risks.

AWS Secrets Manager has added native support for publishing events to Amazon EventBridge whenever a secret value changes. This replaces the previous method of relying on parsed CloudTrail events, which required matching multiple API calls like rotation success or PutSecretValue. The new direct integration allows for real-time, event-driven workflows that can trigger targets such as Lambda, SNS, SQS, or Step Functions upon secret updates.

• Eliminates the need to parse multiple CloudTrail API events to detect secret changes.
• Enables direct EventBridge rules to trigger on active secret value updates.
• Supports real-time workflows via Lambda, SNS, SQS, or Step Functions.
• Simplifies proactive credential caching and rotation handling strategies.

Cloudflare has introduced new capabilities allowing customers to direct AI traffic through its network. These options provide greater control over how AI workloads are handled and routed. The announcement positions Cloudflare to better support the growing demands of AI-driven applications.

• New traffic routing options specifically target AI workloads
• Customers gain more control over AI traffic handling
• Cloudflare aims to support growing AI application demands

Amazon SageMaker AI inference now supports G7e instances in Asia Pacific (Seoul, Tokyo) and Europe (London). These instances deliver up to 2.3x performance over G6e, featuring up to 8 NVIDIA RTX PRO 6000 Blackwell GPUs and 1,600 Gbps networking. The expansion allows low-latency deployment of medium-to-large language models up to 70B parameters closer to end users.

• G7e instances reduce inference latency for GenAI workloads in Asia and Europe.
• Performance is up to 2.3x faster than previous G6e generation instances.
• Each instance supports up to 768 GB total GPU memory for large models.
• Enables serving language models up to 70B parameters locally in new regions.

Amazon SageMaker AI now supports G7 instances powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, delivering up to 4.6x AI inference performance over G6. These instances offer 32 GB GPU memory per card, 5th Generation Tensor Cores, and networking speeds up to 700 Gbps with EFA. The addition of up to 7.6 TB of local NVMe storage allows large models to remain close to compute without over-provisioning.

• Deploy generative AI models with 4.6x higher inference performance compared to G6 instances.
• Access 700 Gbps EFA-enabled networking for faster multi-node training and inference.
• Store large models locally using up to 7.6 TB NVMe SSDs to reduce latency.
• Avoid model quantization and over-provisioning with 32 GB GPU memory per device.

Amazon EKS has added support for Elastic Fabric Adapter (EFA) and EC2 placement groups within EKS Auto Mode and Karpenter node pools. This update allows users to optimize distributed training and inference workloads by controlling instance physical distribution and network interface configurations. Administrators can now specify EFA-only or standard ENI settings on EFA-capable instances, offering finer control over VPC IP utilization.

• EKS Auto Mode and Karpenter now natively support EFA and placement groups.
• Optimize distributed training and inference workloads with better network control.
• Configure EFA-only interfaces to conserve VPC IP addresses on EFA instances.
• Control physical instance distribution across AWS infrastructure for availability.