LodeHQ
AI news for people who shipFree, by email

DevOps · Friday, 29 May 2026 · Issue 4 · 16 stories

DevOps: Harmont, PikoCI, GitHub Actions outage

CI/CD & Automation

PikoCI Offers Single‑Binary Self‑Hosted CI/CD Inspired by Concourse

PikoCI is a lightweight, self‑hosted CI/CD platform shipped as a single binary and inspired by Concourse. It runs anywhere without Docker or Kubernetes, offering a Terraform‑style HCL pipeline language, pluggable back‑ends for storage, queues, and secrets, and supports distributed workers for scaling. The source and demo pipeline are publicly viewable.

pikoci.com ↗2 min read at source

Containers & Orchestration

Nvidia Dynamo Snapshot Cuts Inference Startup to Seconds on Kubernetes

Nvidia’s Dynamo Snapshot combines CRIU and CUDA checkpointing to checkpoint and restore GPU‑accelerated inference containers on Kubernetes, shrinking cold‑start times from minutes to seconds. The prototype restores a full vLLM worker on a new node with near‑instant readiness, helping avoid SLA breaches during traffic spikes. The approach is detailed in an official Nvidia developer blog.

developer.nvidia.com ↗12 min read at source

Observability & Reliability

Adaptive Hedged Requests Slash P99 Latency by 74% in Microservices

Adaptive hedged requests automatically issue parallel backup calls when a request becomes a straggler, cutting overall p99 latency by 74% in microservice fan‑out architectures. The technique uses DDSketch for real‑time latency quantiles and a token‑bucket budget to avoid load spikes during outages, providing a zero‑configuration solution for tail‑latency reduction.

infoq.com ↗17 min read at source

Excess Observability Data Is Raising MTTR and Cloud‑Team Fatigue

A new analysis shows that engineering teams' mean time to resolution has risen annually despite record spending on observability tools, with tool counts climbing to eight‑nine per team. The article argues that beyond a threshold, abundant telemetry creates cognitive overload, masking simple root causes and extending incident recovery time.

devops.com ↗7 min read at source

LinkedIn Leverages eBPF Profiling to Uncover Kernel Lock Contention Freeze

LinkedIn engineers faced recurring 10‑15 second database outages with no logs. By deploying an automated eBPF off‑CPU profiler that captured kernel stack traces at freeze onset, they identified a massive 3.5 GB allocation that blocked the kernel’s mmap_lock, throttling all threads. Pre‑allocating the offending Rust HashMap eliminated the issue.

infoq.com ↗2 min read at source

Cloud & Platform Engineering

DevSecOps

Dirty Frag kernel bug enables container escape and host privilege escalation

The newly disclosed Linux kernel zero‑day (CVE‑2026‑43284 & CVE‑2026‑43500), dubbed “Dirty Frag”, lets a local user in a container gain root on the host and potentially escape isolation. Ubuntu’s advisory details mitigations that disable the vulnerable ESP and RxRPC modules until patched kernels are released. The flaw scores 8.8 CVSS, raising urgent DevSecOps concerns.

ubuntu.com ↗4 min read at source

RHEL 10.2 adds post‑quantum SSH key exchange with FIPS‑compatible hybrids

Red Hat Enterprise Linux 10.2 introduces hybrid post‑quantum key exchange algorithms for SSH, including FIPS‑approved mlkem768nistp256‑sha256 and mlkem1024nistp384‑sha384, plus mlkem768x25519‑sha256 as the default. The updates also bring post‑quantum support to libssh, letting custom SSH services upgrade for "harvest‑now, decrypt‑later" protection.

redhat.com ↗4 min read at source

You’re caught up.

That was all 16 stories in the 29 May issue.