Monthly summary for DataDog/kata-containers — December 2025 Key features delivered - GPU test optimization and runtime class alignment (commits: c505afb67c9ad09b58cd652b33474eebe78b51da; 5b6a2d25bcb6b493c67aa04d9f3518f551900518; 71f78cc87e6b3234fc1a3744789f6695cb23ceac; 50b853eb934c98e401a1622de2a9f18436159dd0). Enable NVIDIA GPU tests as required, reduce memory overhead for GPU runtime classes, lower per-pod memory usage for GPU tests, and align NVIDIA tests with the kata default runtime class. - CI/test infrastructure improvements (commits: 5415cf4e0f20674fd8aa39134de05694036bfa38; 46c7d6c9f85f7a8c02e5e293fe30846af7f3cb56; 69a0ac979ce5d426fdedffb8545a4ae07775ea3a; 3db7b88efff4e386a919c2df38a14cf401049509). Improve CI payload, skip arm64-non-k8s tests, adjust test installer (install_bats), and remove containerd guest pull stability tests. - Release and packaging updates (commits: c7d0c270ee7dfaa6d978e6e07b99dabdaf2b9fda; ded6d1636f29d8d36a9cbe6aee8946b4debb5bfe; 1388a3acdad721977da7368052782555b2974542; a25a53c8602a7a27d92166d3b41183dded310092). Bump version to 3.24.0, remove deprecated features from 3.23.0, add ORAS cache for gperf and busybox tarballs, and fix permissions for patching nodefeaturerules. - NVIDIA runtime improvements (commits: 995770dbeb3f8b0d5c18daf64a0b6de712704602; 88cdfab604c108a8667c1d956e4234b987e0d541). Use cold-plug by default and align static_sandbox_resource_mgmt. - Kata-remote runtime class support with Helm (commits: 6e01ee6d477680e81145b2af2fbe985208ad9710; 35cd5fb1d4763dc85728b4f916d74119320bf6ef). Add kata-remote runtime class support and update Helm to v4.0.4. Major bugs fixed - GPU rootfs handling fix: Temporarily revert GPU-related root_hash.txt handling in rootfs to restore stable behavior. (commit 923f97bc6658bad1ccdb33b7188b992e171ca5e0) - Revert CDI files workaround: Revert the CDI files workaround added for tests. (commit 406f6b1d157d8e65660a8e66707fc34388993915) - Build system: Fix GPG key for gperf. (commit b11cea31138349f4436b7902c78d7be2f73c381d) Overall impact and accomplishments - Delivered measurable improvements in test efficiency and resource usage through GPU-optimized tests and memory footprint reductions, resulting in faster CI cycles and lower operational costs. Strengthened release engineering with a 3.24.0 upgrade, ORAS caching, and packaging reliability. Enabled scalable enterprise deployments via NVIDIA runtime improvements and Helm-based kata-remote runtime class support. Technologies/skills demonstrated - Kubernetes and Kata Containers runtime configurations; NVIDIA GPU runtime optimizations; CI/CD optimization and test stability; Helm and Rust (kata-deploy); ORAS-based caching; release engineering and packaging; build security (GPG) and version management.

November 2025 (2025-11) monthly summary for DataDog/kata-containers focused on delivering deployment reliability, test stability, and CI improvements. Key platform enhancements include Helm-driven RuntimeClass creation (with arch-specific annotations) and chart-driven runtimeClass tests; structured configuration for shims with backward compatibility and accompanying example values; Rust-based kata-deploy integration into Helm and CI, plus a nightly CI job to validate Rust deployment paths. In parallel, Kubernetes/NVIDIA GPU test suites were stabilized through larger-instance QoS tests, extended parallel test timeouts, and targeted stability tests for experimental-force-guest-pull, with several test-name refinements. CI and scripting were enhanced via per-shim proxy handling, deduplication safeguards, and a Go upgrade, driving more reliable releases and faster feedback cycles.

October 2025: NVIDIA/kata-containers focused on improving CI reliability for the Initramfs build. Implemented a two-step process to generate and compress the initramfs archive, isolating failures and providing clearer error reporting. This change reduces CI flakiness, accelerates debugging, and supports more stable nightly builds.

September 2025 — DataDog/kata-containers: Delivered CI-focused enhancements for containerd integration, enabling earlier issue detection and safer pre-release validation. No major customer-facing bugs fixed this month; instead, reliability of the containerd integration was improved and risk of regressions reduced through proactive testing. Overall impact: faster feedback, stronger release confidence, and reduced deployment risk due to broader pre-release testing in CI. Key improvements delivered: - Containerd CI testing enhancements: updated versions.yaml to set active v2.1 and latest v2.2 - Relaxed CI/install tests to allow unstable (pre-release) containerd releases - Early issue detection enabled by broader pre-release validation in CI Note: Changes are tracked in DataDog/kata-containers; commits included aa9e3fc3d5aa17ea86e94276cd11d9c527b706f7 and 287db1865fcefb3481c218d0d5e31fd124f156c0

July 2025 focused on reliability and maintainability of kata-containers deployments. Implemented kata-deploy Helm uninstall enhancements to align uninstall flow and added nodeSelector support for uninstall cleanup jobs, delivering more stable tests and cleaner cleanup.