Break point: Cycloid, Istio, Sumo Logic, FastAI.jl, O3DE, PyTorch Lightning, and Linkerd

DevOps tooling company Cycloid launched infrastructure-as-code product Infra Import into beta this week. The tool is based on the company’s open-source project TerraCognita and was developed to deduce Terraform files and related Git modules from manually provisioned cloud deployments. The resulting artefacts can then be used for version control and automation, keeping setups reproducible and easier to integrate into compliance measures.

Istio 1.9 is nearing its end of life

The Istio team reminds users of version 1.9 of the service mesh to get updating, since it will stop back-porting security and critical fixes for the minor release on 18 August, 2021. The most recent version available is v1.10.3, which landed about two weeks ago. 

Besides keeping your setup safe, switching to the 1.10.x series brings improvements such as stable revision labels, discovery selectors to limit the set of resources Istiod should watch, and less trouble when working with Kubernetes’s StatefulSets.

Sumo Logic bulks up on big brother capabilities

Monitoring and analytics purveyor Sumo Logic updated its platform, adding span analytics and real user monitoring (RUM) into the mix. While the former lets teams query and analyse application data, RUM is meant to provide information about how users experience a specific web application so that performance issues and similar troubles can be mitigated more quickly. More precisely, the feature collects “details about the end-user geolocation, device and browser, including all details about the URL, target element and its Xpath indicating the particular component of the page that was clicked”.

Julia language gets deep learning best practices repo

After over one year in the works, there’s finally a Julia version of the popular Python deep learning framework fast.ai. FastAI.jl uses the Flux machine learning library and looks to provide Julia programmers with a user-friendly way to create deep learning models. Library features are said to include a learning rate finder, efficient data augmentation and loading, visualisations, logging capabilities, and “high-level training loops with hyperparameter scheduling and callbacks”.

Developers interested in the project should put a blocker into their calendar, since the project initiators will get together with fast.ai co-creator Jeremy Howard on 2nd August to answer questions around the library. A thorough comparison of both tools can meanwhile be found in the FastAI.jl documentation.

Open source devs make O3DE engine ready for Linux

The O3DE project just showed that sometimes open sourcing really helps in closing availability gaps. After AWS handed the 3D engine over to the Linux Foundation just a couple of weeks ago, the project has a working Linux editor and includes some other patches to forward Linux support. O3DE is based on Amazon’s Lumberyard engine and served as the foundational project for the newly announced Open 3D Foundation — a move some onlookers found puzzling, given that Linux wasn’t even among the officially supported platform.

PyTorch Lightning learns to work with TPU pods

PyTorch research framework Lightning has been released in version 1.4. The update introduces support for TPU pods and Graphcore’s Intelligence Processing Units as well as a XLA profiler for debugging cloud TPU training workload performance. Additions in beta include plugins to allow fully sharded parallelism and fine-tuning very large numbers of parameters via NVMe offloading.

Linkerd finally graduates CNCF

The Cloud Native Computing Foundation just welcomed a new project into its graduated project roaster. Service mesh Linkerd was the fifth project donated to the foundation back in 2017, and the first to join its sandbox. The new label is meant to signal stability and maturity to potential users, but also stands for additional support from the CNCF.

Linkerd isn’t the only service mesh in the cloud native space (see Istio), but stands out for its focus on simplicity and usability. Current plans for the project are said to include server- and client-side policies, as well as “mesh expansion” so that the Linkerd data plane can operate outside of CNCF flagship Kubernetes.