TLDR DevOps 2026-04-10

Pulumi now supports Bun as a fully supported runtime (not just a package manager) as of version 3.227.0, allowing developers to execute entire Pulumi programs by setting `runtime: bun` in their Pulumi.yaml with no Node.js required. The integration offers faster package installation and native TypeScript support, though it currently doesn't support policy packs, function serialization, or multi-language components that some projects may still need Node.js for.

Databricks launched public preview support for Iceberg v3, which introduces Row Lineage for tracking data changes, Deletion Vectors for 10x faster updates and deletes, and a VARIANT type for querying semi-structured data without schema enforcement. The update eliminates the performance tradeoff between Delta Lake and Iceberg through Unity Catalog's cross-engine governance, allowing teams to write data once and read it across multiple platforms like Snowflake, BigQuery, and Redshift without replication.

Amazon S3 Files is a new capability that allows any S3 bucket to be mounted and accessed as a fully-featured file system directly from EC2 instances, containers, or Lambda functions, with ~1ms latencies for active data and automatic bidirectional synchronization between the file system and S3 bucket. The service is now available in all commercial AWS Regions and eliminates the traditional tradeoff between S3's object storage benefits and the interactive capabilities of file systems, making it particularly useful for collaborative workloads like AI agents and ML training pipelines that need shared, low-latency access to S3 data.

Traceroute works by sending packets with increasing TTL values so each router drops them in sequence and returns an ICMP “Time Exceeded” message, revealing each hop's IP until the destination responds with a different ICMP message. By repeating this process with multiple probes and measuring response times, it maps the path and latency of network traffic, though gaps occur when routers don't send replies.

Ingress NGINX reached end-of-life in March, forcing organizations to urgently migrate to Kubernetes Gateway API due to critical vulnerabilities like CVE-2026-24512 and CVE-2026-3288 that now have no supported fixes. This article provides a technical migration guide that covers controller selection, baseline performance tracking, parallel deployment strategies, and DNS cutover procedures to help teams safely transition away from the deprecated ingress controller.

“Why is my spend 5% higher this month?” Noros is the AI agent that answers in seconds. Ask about anomalies, overprovisioned instances, cost spikes, and more across AWS, GCP, and Azure.

TLDR DevOps readers get a free month at launch. Join the waitlist →

Tailslayer is a C++ library that reduces tail latency in RAM reads caused by DRAM refresh stalls.

Veralang is a programming language designed for LLMs to write, not humans. Vera's principles include prioritizing explicitness over convenience and checkability over correctness.

Archon is the first open-source workflow engine for AI coding agents. It allows developers to define development processes like planning, testing, and PR creation as deterministic YAML workflows rather than relying on unpredictable AI behavior. The tool, which works with platforms like Claude Code, includes 17 default workflows, a web dashboard for monitoring, and aims to bring the same standardization to AI coding that Dockerfiles brought to infrastructure.

Netflix reduced massive redundant query load in Apache Druid by introducing interval-aware caching that reuses previously computed time-bucketed results and only queries fresh data, using exponential TTLs that reflect data freshness. This approach trades minimal staleness for significant efficiency gains, cutting query volume, improving latency, and shifting scaling from expensive compute to cheaper cache infrastructure.

Meta avoided the “forking trap” by replacing a divergent WebRTC fork with a dual-stack architecture using a shim layer that lets legacy and upstream versions run side-by-side, enabling safe A/B testing and continuous upgrades. Combined with a feature-branch system for managing patches, this approach improved performance, reduced binary size, enhanced security, and allowed ongoing alignment with upstream releases without repeated costly migrations.

Turn incidents into learnings in minutes, not hours. incident.io now uses AI to scan threads and PRs, automatically generating data-rich postmortem drafts and live-synced incident timelines. Learn more.

PostHog fixed major latency spikes by identifying that CPU-bound Rayon tasks were blocking Tokio's async I/O threads, causing scheduler throttling and application freezes rather than a database issue.

AI-generated code should be transparently disclosed using a standardized AI-DECLARATION.md file that clearly states how much AI was involved across levels ranging from none to fully autonomous (auto), optionally broken down by development phases or components.