TLDR DevOps 2026-06-29

Know what your AI is actually doing? At New Relic NOW, you'll learn to consolidate tooling, govern non-deterministic architectures, and ship reliable AI-native apps, all in one morning.

Topics include:

✅ Real lessons from engineering leaders at OpenAI and LaunchDarkly.

✅ First look at exclusive data on how AI-generated code and prod failures, rework time, and observability policies

✅ How to embed observability into your AI dev loop

Save your spot

Google Cloud launched SQL-based alerting in Observability Analytics (in preview), allowing developers to create complex alerts by writing SQL queries over logs and traces instead of relying on simple threshold monitoring. The feature enables advanced use cases like alerting on error rate percentages for specific customers or tracking p99 latency across high-cardinality data, with queries running on scheduled intervals through BigQuery and triggering notifications via channels like Slack or PagerDuty.

Grafana Cloud's Kubernetes Monitoring app now provisions alerts using Grafana-managed alerting instead of data source-managed alerts, which can cause notification routing issues after reinstallations because the two systems use separate contact points, policies, and Alertmanagers. Users should upgrade via the app's Update button, reconfigure notifications in Grafana's built-in alerting system after reinstalls, and migrate custom alert rules out of the integrations-kubernetes namespace to avoid losing them during upgrades.

VictoriaLogs stores logs as immutable, searchable parts grouped by stream and partitioned by day, making retention cheap, queries time-bounded, and recent data searchable before it fully lands on disk. Its columnar layout lets queries read only the fields they need, while stable low-cardinality stream fields, per-day partitions, metadata pruning, and compression help operators understand why some log queries stay fast and others get expensive.

Semgrep benchmarked open-weight and frontier models on IDOR vulnerability detection and found that GLM 5.2, using only a minimal prompt in a Pydantic AI harness, outperformed Claude Code on this specific security task at much lower cost. The results reinforce that harness design still matters most, but open-weight models are becoming credible options for security teams that need cheaper, private, and swappable model infrastructure.

Strix, an open-source security testing tool, uses autonomous AI agents to dynamically scan code for vulnerabilities and generate proof-of-concepts, integrating directly into GitHub Actions to automatically block insecure code before production. The tool addresses common pentesting pain points by running actual exploit validation rather than static analysis, with results automatically scoped to changed files in pull requests and support for enterprise deployments with SSO and custom compliance options.

CuPy, a NumPy/SciPy-compatible array library developed by Preferred Networks, acts as a drop-in replacement to run existing Python scientific computing code on NVIDIA CUDA or AMD ROCm GPU platforms. The open-source library is available via PyPI and Conda-Forge, with version 13.0.0 incorporating cuSignal, and offers access to low-level CUDA features including RawKernels and direct CUDA Runtime API calls.

Octopus Deploy made ephemeral environments optional for pull requests by using a GitHub label that conditionally skips deployment jobs while preserving branch protection through a gate job that validates required workflow outcomes. The approach reduces unnecessary deployment time and infrastructure costs for low risk changes while allowing teams to share a centralized workflow improvement that benefits all repositories using the platform.

Flink 2.3 adds flink-s3-fs-native, an experimental opt-in S3 filesystem that removes Hadoop dependencies, moves to AWS SDK v2, supports exactly-once sinks, and gives Flink one S3 plugin for checkpoints, savepoints, and file sinks. Production-scale benchmarks showed roughly 2× faster checkpoints, smaller checkpoint storage, cleaner s3.* configuration, and a safer migration path for teams running stateful Flink jobs on S3.

Memory access order can make a simple integer summation dramatically slower by defeating cache locality, prefetching, and page-level memory behavior. Sequential, random, and deliberately adversarial layouts show why CPU performance depends heavily on how data is arranged and accessed, not just on the number of operations executed.

How do you deal with slow CI pipelines delaying delivery? Global financial institution ING accelerated its iOS build pipeline for 60+ engineers by running macOS VMs on Orka. Read the story

At New Relic NOW 2026, the company introduced Autopilot for autonomous incident investigation, Ground Truth for AI-optimized access to observability data, expanded AI observability capabilities, new compliance commitments, and startup offerings aimed at improving reliability, governance, and operational efficiency in AI-first software development.

Amazon EKS now supports customer routed control plane egress, allowing Kubernetes API Server outbound traffic, such as admission webhooks, OIDC lookups, and aggregate API requests, to flow through customer VPCs with existing routing, security, firewall, and compliance controls.

The Kubernetes Security Profiles Operator hit v1.0.0, its first stable release after six years of development, graduating all eight APIs and passing a third-party security audit with zero critical vulnerabilities found.