Part of IT Trends & Reviews — what actually shipped, quarter by quarter.
1. Introduction: The Cloud-Native Inflection Point
Q3 2017 — the three months from July through September — is a quarter that information-technology professionals will study for years. In fact, several distinct storylines collide simultaneously, creating a moment of unusual intensity. For example, serverless computing crosses the threshold from early-adopter experiment to an engineering discipline that enterprises are actively piloting. Meanwhile, Kubernetes releases version 1.8, cementing its status as the default substrate for containerised workloads and absorbing the container-orchestration wars for good. And on September 26, React 16 ships — with a full MIT licence — ending weeks of community drama that had threatened to fragment the JavaScript ecosystem.
Moreover, the quarter’s security dimension is equally significant. For example, WannaCry ransomware struck in May and NotPetya in June, leaving organisations still conducting post-mortems as July opens. Then, in early September, Equifax discloses a breach affecting approximately 145 million Americans — exposing a catastrophic failure in patch management. The breach, caused by an unpatched Apache Struts vulnerability known since March, transforms the abstract fear of technical debt into a concrete boardroom imperative. Security, finally, demands the same engineering rigour as reliability.
1.1 Blockchain, AI, and the quarter’s three crises
Meanwhile, blockchain excitement reaches a fever pitch. Notably, the Enterprise Ethereum Alliance — which formed in March with 30 founding members — now counts over 116 enterprise members including Microsoft, Intel, and J.P. Morgan. Meanwhile, initial coin offerings are exploding. By comparison, Ethereum itself trades in a range of roughly $170 to $400 during the quarter, far below the stratospheric peaks that arrive in late 2017 and early 2018. However, already a dramatic run from single-digit prices eighteen months earlier. Overall, the ICO boom raises both genuine excitement and equally genuine concern about speculation and regulatory clarity.
Meanwhile, on the AI front, Google Brain’s landmark “Attention Is All You Need” paper — published just before Q3 opens — is circulating widely in machine-learning circles. In addition, TensorFlow 1.3 ships in August, adding Keras as a high-level API that dramatically lowers the barrier to deep-learning experimentation. Similarly, Apple announces the iPhone X on September 12. Notably, it features the A11 Bionic chip with a dedicated Neural Engine. This is the first time a mass-market smartphone processor integrates a specialised AI accelerator. Overall, the hardware story and the software story of AI are beginning to converge in consumers’ pockets.
Key Insight: Three Crises, One Quarter
Overall, Q3 2017 is defined by three simultaneous crises with structural consequences. First, the React licensing controversy forces the entire open-source community to examine how corporate patent clauses interact with permissive licences. Second, the Equifax breach proves that patch management is a life-or-death operational requirement, not optional hygiene. Third, the ICO mania building around Ethereum tests whether blockchain technology can separate genuine utility from speculative frenzy. Overall, each crisis accelerates the professional maturity of its domain.
2. GitHub Deep Dive — Key Repositories of Q3 2017
The open-source activity of Q3 2017 concentrates around five distinct technology narratives: container orchestration (Kubernetes), UI rendering (React), serverless frameworks (Serverless Framework, OpenFaaS), machine learning (TensorFlow), and GraphQL (Apollo Client). In practice, each of the repositories below shapes how teams architect systems during this quarter and well into 2018.
2.1 Kubernetes 1.8 — Security, Workloads, and Feature Depth
kubernetes/kubernetes — v1.8.0
First, Kubernetes 1.8 arrives at the end of Q3 with a theme the release team labels “Security, Workloads and Feature Depth”. The headline change is Role-Based Access Control (RBAC) graduating to stable. Notably, that is a critical milestone. It finally allows enterprises to enforce least-privilege access policies at the cluster level. In practice, before RBAC stability, production deployments frequently relied on permissive cluster-admin bindings, a configuration that security teams rightly flag as untenable.
Additionally, 1.8 delivers Horizontal Pod Autoscaler v2 graduating to beta, enabling autoscaling on custom metrics — not just CPU — through the new metrics API. For example, teams can now scale deployments based on queue depth, request latency, or any metric surfaced via Prometheus. Network policy improvements extend Kubernetes to specify egress rules in addition to ingress, giving platform teams fine-grained control over pod-to-pod communication. In addition, volume snapshots enter alpha, previewing a future where stateful workloads gain portable backup primitives directly integrated with the Kubernetes API.
Parallel to the 1.8 release, VMware announces Kubernetes integration in its Pivotal Container Service (PKS) offering in August. Similarly, Mesosphere, maker of DC/OS, announces Kubernetes support in September. As a result, within a single quarter, two major infrastructure vendors effectively abandon competing orchestration approaches to build on Kubernetes. In short, the container-orchestration wars are over.
2.2 React 16 — Fiber, MIT Licence, and the End of Patent Drama
facebook/react — v16.0.0
Meanwhile, React 16 is arguably the most anticipated JavaScript release of 2017 — and not solely for technical reasons. First, the quarter begins with a licensing crisis. For example, Facebook ships React under the BSD+Patents licence, a non-standard arrangement the Apache Software Foundation declares incompatible with its policies. Meanwhile, on September 15, WordPress announces it will drop React from its Gutenberg editor project, citing the patent clause as an unacceptable legal risk for downstream users. The WordPress decision sends shockwaves through the ecosystem. In fact, it signals that a major production user is willing to absorb the engineering cost of migration. In fact, tens of millions of sites depend on its software. That user would rather move away from React than accept the licence terms.
Therefore, Facebook responds rapidly. In addition, on September 23, the company announces it will relicence React, Jest, Flow, and Immutable.js under the standard MIT licence. Finally, three days later, React 16.0.0 ships with the MIT licence applied. The crisis is resolved, and React emerges stronger. Notably, the swift resolution demonstrates Facebook’s commitment to the open-source community. It also removes the single largest adoption barrier for enterprise users.
Technically, React 16 is a ground-up rewrite of the internal rendering engine, dubbed Fiber. By contrast, the old Stack renderer processed the entire component tree synchronously, causing janky animations in complex UIs. Instead, Fiber breaks rendering into incremental units that can be paused, resumed, and prioritised by urgency. Notably, the user-visible API is entirely unchanged, so existing applications upgrade without code modifications. Additionally, React 16 introduces error boundaries — class components implementing componentDidCatch — which prevent a JavaScript error in one subtree from crashing the entire application.
2.3 Serverless Framework — Multi-Cloud, Version 1.x Stable
serverless/serverless — v1.19–v1.22
Meanwhile, the Serverless Framework reaches maturation in Q3 2017 as the de facto standard for defining and deploying serverless functions across cloud providers. Its serverless.yml configuration format becomes a common dialect: teams express function triggers, environment variables, IAM permissions, and API Gateway routes in a provider-agnostic schema, then let the framework generate provider-specific infrastructure templates. AWS Lambda, Azure Functions, Google Cloud Functions, and IBM OpenWhisk all support deployment through a unified CLI.
Notably, the 1.19–1.22 release cycle, landing throughout Q3, focuses on plugin stability and local development experience. The serverless invoke local command simulates Lambda execution locally, dramatically accelerating the inner development loop for teams who previously had to deploy to a live cloud environment to test their functions. The plugin ecosystem grows to over 500 community-contributed extensions, covering everything from webpack bundling to DynamoDB local emulation.
2.4 OpenFaaS — Serverless on Kubernetes
openfaas/faas — Functions as a Service
OpenFaaS, created by Alex Ellis, moves to the openfaas GitHub organisation in Q3 2017 as community interest accelerates. The project takes a fundamentally different approach from AWS Lambda. Rather than depending on a proprietary cloud service, OpenFaaS packages every function as a standard Docker container. It then runs that container on any Kubernetes or Docker Swarm cluster. This means teams can deploy serverless workloads on-premise, on bare metal, or on any cloud — including private infrastructure with no public-cloud connectivity.
Free ebook
Free AI Video, Generated Locally
Working scripts and measured benchmarks. Free.
No spam. Unsubscribe at any time.
The project attracts thousands of GitHub stars during Q3. The “serverless on Kubernetes” narrative resonates with operators. They want the developer experience of functions without surrendering infrastructure portability. The faas-cli tool allows developers to build, push, and deploy functions in a single command. OpenFaaS ships with a Prometheus-based monitoring stack, giving teams immediate visibility into function invocation rates, durations, and error rates.
2.5 TensorFlow 1.3 — Keras Arrives
tensorflow/tensorflow — v1.3.0
TensorFlow 1.3 ships in August with a change that has outsized consequences for AI accessibility: Keras is included as tf.contrib.keras, making the popular high-level neural network API an official part of the TensorFlow ecosystem. Before 1.3, building a convolutional neural network in TensorFlow requires writing boilerplate session management, variable initialisation, and graph construction code that obscures the actual model architecture. Keras provides a clean, layer-oriented API that reduces this cognitive overhead dramatically.
TensorFlow 1.3 also improves XLA (Accelerated Linear Algebra) compilation for TPU hardware. The release coincides with growing institutional adoption: universities integrate TensorFlow into graduate machine-learning curricula, and enterprise teams begin building production inference pipelines on TensorFlow Serving. Critically, TensorFlow 1.3 runs on the newly announced Google Cloud TPU in Alpha — the first time developers can access specialised AI silicon without purchasing hardware outright.
2.6 Apollo Client — GraphQL Goes Mainstream
apollographql/apollo-client — 2.0 Beta
Apollo Client 2.0 is in active beta development throughout Q3 2017, with a stable release landing in November. The 2.0 architecture introduces a radical simplification: the optional dependency on Redux is removed, replaced by a purpose-built in-memory cache. This reduces bundle size, eliminates boilerplate, and makes Apollo easier to reason about in large codebases. The Apollo Link system replaces network interfaces with a composable middleware chain. That chain lets teams inject authentication headers, implement retry logic, or route queries to different endpoints. None of this requires modifying application code.
Apollo’s momentum in Q3 2017 reflects GraphQL’s broader trajectory. Facebook open-sources the GraphQL specification in 2015. In addition, by mid-2017, GitHub publicly announces it is rewriting its public API using GraphQL — a credibility signal that accelerates enterprise adoption significantly.
3. Big Tech & Industry Breakthroughs
Q3 2017 sees major technology companies make strategic moves that define competitive positions for years. The table below captures the most significant announcements, acquisitions, and product launches from the global technology industry during July–September 2017.
| Company | Event | Significance | Region |
|---|---|---|---|
| Apple | Announces iPhone X and A11 Bionic on September 12, 2017 | iPhone X introduces Face ID, OLED Super Retina display. In addition, the A11 Bionic chip with a dedicated Neural Engine capable of 600 billion operations per second — the first mass-market smartphone chip with specialised AI inference silicon. | USA |
| Relicences React, Jest, Flow, and Immutable.js under MIT on September 23 | Ends the BSD+Patents licensing controversy that threatened enterprise and open-source adoption. The decisive week-long response to community pressure demonstrates how quickly corporate maintainers can act when business risk is sufficiently clear. | USA | |
| Cloud TPU Alpha opens to researchers via Google Cloud | The Tensor Processing Unit — Google’s custom ASIC for neural network inference — is available to external researchers for the first time. This marks the beginning of specialised AI hardware becoming accessible beyond hyperscalers’ internal infrastructure. | USA | |
| Google / Google Brain | “Attention Is All You Need” paper (June 2017, widely circulated Q3) | Vaswani et al. introduce the Transformer architecture. The paper is the foundation for GPT, BERT, and every major language model of the next decade. In Q3 2017, researchers are rapidly replicating and extending the results. | USA |
| AWS | Lambda surpasses 10B+ invocations/month; Step Functions expand | AWS Lambda’s scale validates the serverless model at enterprise volume. AWS Step Functions expands capabilities, allowing long-running state machines that coordinate Lambda functions, ECS tasks, and other AWS services without managing infrastructure. | USA |
| Microsoft | Azure Kubernetes Service (AKS) preview announced | Microsoft enters the managed Kubernetes market with AKS, a fully managed control plane offering that competes with Google GKE. Every major cloud provider now commits to supporting Kubernetes as managed infrastructure. | USA |
| VMware | Announces Kubernetes support in Pivotal Container Service (PKS), August 2017 | VMware — the dominant enterprise virtualisation vendor — embraces Kubernetes rather than competing with it. The VMware endorsement is a critical moment for enterprise Kubernetes adoption in on-premise data centres. | USA |
| Mesosphere | Adds Kubernetes to DC/OS alongside Marathon, September 2017 | Mesosphere integrates Kubernetes as a first-class scheduling framework in DC/OS, effectively ending the DC/OS vs. Kubernetes debate for most enterprise workloads. | USA |
| Equifax | Discloses breach of approximately 145 million records, September 7, 2017 | Caused by an unpatched Apache Struts vulnerability (CVE-2017-5638) known since March 2017. The largest consumer credit bureau breach in history turns security debt from a theoretical risk into front-page news and Congressional testimony. | USA |
| Alibaba Cloud | Launches Container Service for Kubernetes (ACK) | Alibaba Cloud’s ACK brings managed Kubernetes to Asia-Pacific markets. Cloud-native container infrastructure is confirmed as a global, not solely Western, phenomenon. | China |
| Tencent Cloud | Tencent Kubernetes Engine (TKE) reaches General Availability | China’s largest internet company operating Kubernetes clusters at WeChat scale provides compelling proof that the technology handles extreme production throughput. TKE reaches GA in 2017 serving both WeChat’s backend and external enterprise customers. | China |
| Spotify | Engineers publish detailed GKE migration case study | Spotify’s public documentation of its migration to Google Kubernetes Engine becomes required reading for platform engineering teams worldwide, demonstrating that Kubernetes handles the scale of a global streaming service with hundreds of millions of users. | Europe |
3.1 The React Licensing Crisis in Detail
The React licensing situation deserves additional context because it reveals structural tensions in how large technology corporations maintain critical open-source infrastructure. Facebook adopts the BSD+Patents licence for React in 2014, adding a clause that terminates the patent grant if the licensee asserts any patent claim against Facebook. The Apache Software Foundation identifies this as incompatible with its policies in July 2017, declaring that software distributed under BSD+Patents cannot be included in Apache-licensed projects.
Facebook’s engineering leadership initially defends the licence choice in August 2017. It argues that the patent termination clause only affects licensees who actively litigate against Facebook. That scenario, they contend, does not threaten ordinary React users. However, the practical consequence is severe. Any startup that uses React risks losing its right to use React in production. The risk appears when it later faces a patent dispute involving Facebook, even an unrelated one.
WordPress’s decision to abandon React is the breaking point. WordPress powers approximately 28% of all websites as of 2017. Its engineering team’s judgment that the licence risk is too significant for a CMS ecosystem with millions of plugin authors carries enormous weight. Within eight days of the WordPress announcement, Facebook relicences React under MIT. The reversal is fast and complete, extending to Jest, Flow, and Immutable.js. That demonstrates that community signals, when sufficiently clear, can change even large-scale corporate open-source policy.
4. AI & Technology Impact
The AI landscape of Q3 2017 sits at an inflection point. Deep learning is producing results that a decade of classical machine-learning research could not match. However, the tooling required to apply it remains intimidating. TensorFlow 1.3’s Keras integration is the most visible step toward democratisation this quarter. Simultaneously, the publication of the Transformer paper and the shipping of Apple’s Neural Engine signal two structural changes. Those changes define the next several years of AI development. They are algorithmic innovation in attention mechanisms and silicon specialisation for inference.
“Attention Is All You Need” — The Transformer Paper
Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, and Polosukhin at Google Brain publish the Transformer architecture. Multi-head self-attention replaces recurrent connections, enabling fully parallelised sequence-to-sequence modelling. The WMT 2014 English-to-French benchmark result — 41.0 BLEU — surpasses all previous single-model results. NLP researchers are rapidly replicating the architecture. In addition, it quickly becomes clear that the Transformer will replace LSTMs and GRUs as the default sequence modelling primitive.
TensorFlow 1.3 — Keras as First-Class Citizen
Google releases TensorFlow 1.3 with Keras included in tf.contrib.keras. The integration validates Keras as the standard high-level API for TensorFlow and provides a migration path for researchers who need TensorFlow’s distributed training infrastructure. XLA improvements deliver measurable speedups on Cloud TPU hardware, which opens to external researchers in alpha during this quarter.
4.1 Frameworks and silicon specialisation
PyTorch 0.2.0 — Distributed Training Enters PyTorch
Facebook AI Research releases PyTorch 0.2.0, adding torch.distributed for data-parallel and model-parallel training across multiple GPUs and multiple machines. PyTorch’s dynamic computation graph — which allows network architectures to change on each forward pass — makes it the preferred research framework for novel architecture experiments. A growing share of NIPS 2017 paper submissions cite PyTorch implementations, signalling a generational shift away from Caffe and Theano.
Apple A11 Bionic — The Neural Engine Enters Consumer Devices
Apple announces the A11 Bionic chip, built on TSMC’s 10nm process, at its iPhone X and iPhone 8 event. A dedicated Neural Engine — a two-core processor optimised for neural network inference — handles Face ID’s face-recognition pipeline, animoji generation, and ARKit augmented-reality features. Capable of 600 billion operations per second, this is the moment when AI inference hardware moves from cloud data centres to billions of consumer devices. The implications for on-device ML — privacy-preserving inference, offline capability, low latency — begin to emerge as engineering priorities.
4.2 Generative models and the two tracks of AI progress
GAN Research Proliferates
Generative Adversarial Networks are producing increasingly convincing synthetic images by Q3 2017. Research groups publish conditional GANs, progressive growing GANs, and domain transfer methods including CycleGAN (March 2017). These results demonstrate that deep networks can synthesise photorealistic content with limited supervision. The quality of synthetic images is improving rapidly enough that discriminating real from generated content is becoming non-trivial for human evaluators.
Key Insight: The Two-Track AI Acceleration
Q3 2017 sees AI progress on two parallel tracks. The algorithmic track — represented by the Transformer paper — produces architectural innovations that will define the next decade of NLP and beyond. The hardware track — represented by the A11 Bionic Neural Engine and the Cloud TPU alpha — democratises the compute required to run inference at scale. These tracks do not converge immediately. However, their simultaneous acceleration in Q3 2017 establishes the structural conditions for the large-model era that follows.
5. Key Voices & Thought Leaders
Several engineers and researchers shape the technical discourse of Q3 2017 through blog posts, conference talks, and open-source contributions. Their perspectives help practitioners navigate the signals from Kubernetes releases, React licensing debates, and serverless adoption curves.
5.1 Kelsey Hightower — Kubernetes Practitioner
Kelsey Hightower, a developer advocate at Google, is the most visible and accessible voice explaining Kubernetes to practitioners in 2017. His “Kubernetes the Hard Way” tutorial is a step-by-step guide. It bootstraps a production-grade Kubernetes cluster entirely from scratch, without automation. It becomes the canonical reference for understanding what the control plane actually does. Hightower’s talks at KubeCon and DockerCon emphasise operational discipline: Kubernetes is a powerful tool that rewards understanding over cargo-culting.
5.2 Dan Abramov — React Core Team
Dan Abramov, creator of Redux and a member of Facebook’s React core team, is the face of React’s public communication during the licensing controversy. He is transparent about the rationale for the MIT relicensing. He acknowledges community concerns, explains internal decision processes, and commits publicly to the change. That transparency is credited with preserving React’s credibility. During Q3 2017, Abramov engages actively on Twitter and GitHub. He addresses technical questions about the Fiber architecture and the error boundary API. Those answers help developers understand and trust the upgrade path to React 16.
5.3 Alex Ellis — OpenFaaS Creator
Alex Ellis, a software architect and Docker Captain, creates OpenFaaS and grows it from a personal project to a recognised open-source community during 2017. His blog posts on blog.alexellis.io document the design decisions behind OpenFaaS. They explain why Docker containers rather than language-specific runtimes are the right abstraction for portable serverless functions. They also cover how to run production workloads on Kubernetes. Ellis’s work attracts contributors from the enterprise Kubernetes community who want serverless capabilities without cloud-provider lock-in.
5.4 Martin Fowler — Microservices and Architecture
Martin Fowler, ThoughtWorks Chief Scientist, continues to be a defining voice on software architecture through his martinfowler.com blog. In Q3 2017, his writing covers the tradeoffs of microservices architectures. It includes the operational complexity that serverless adoption must navigate. That writing provides the conceptual vocabulary used by architecture review boards and engineering managers worldwide. Fowler’s collaboration with Sam Newman on microservices patterns gives teams a shared language for discussing service decomposition, event-driven coordination, and distributed data management.
5.5 Charity Majors — Observability and Distributed Systems
Charity Majors, co-founder and CTO of Honeycomb, pushes the observability narrative during Q3 2017. She argues that traditional monitoring dashboards are insufficient for debugging distributed microservices and serverless applications. Her writing covers the need for high-cardinality event data. It also draws a distinction between monitoring (aggregating metrics) and observability (exploring raw events). That work aligns directly with the operational challenges that Kubernetes and serverless adoption are creating. As more teams operate at a layer of abstraction where traditional APM tools struggle, Majors’s frameworks for thinking about production systems gain traction among platform engineering practitioners.
6. Trend Synthesis
Standing at the end of September 2017 and looking across the quarter’s signals, several structural shifts are unmistakably underway. The convergence of these shifts is not coincidental — they reinforce each other through shared technical infrastructure and shared community norms.
The first and clearest signal is the consolidation of Kubernetes as the standard unit of infrastructure composition. Kubernetes 1.8’s RBAC graduation, VMware’s PKS announcement, and Mesosphere’s DC/OS integration mark a phase transition. So do the preview launches of managed Kubernetes from Microsoft (AKS), Alibaba Cloud (ACK), and Tencent Cloud (TKE). Kubernetes is no longer a technology that organisations evaluate against alternatives — it is the substrate on top of which alternative approaches are now built. OpenFaaS running on Kubernetes. Service meshes running on Kubernetes. Streaming platforms running on Kubernetes. The container-orchestration debate is settled.
The second signal is the maturation of the serverless execution model. AWS Lambda’s scale — exceeding 10 billion invocations per month — proves that the function-as-a-service model handles production traffic at enterprise volumes. The Serverless Framework’s multi-cloud deployment model and OpenFaaS’s Kubernetes-native approach represent two architectural poles: cloud-managed simplicity versus infrastructure portability. Both attract real production workloads in Q3 2017. That suggests “serverless” is not a monolithic category but a spectrum of execution models. Those models are united by the principle that function authors should not manage server provisioning.
6.1 Open-source governance as the third signal
The third signal — and perhaps the most consequential for long-term ecosystem health — is the open-source licensing wake-up call. The React controversy is not primarily a story about Facebook’s patent strategy. It is a story about the assumptions underlying the open-source social contract. When a project of React’s scale ships under a non-standard licence variant, the disruption extends far beyond the immediate technical community. Legal teams at financial institutions, hospital systems, and government agencies must review their dependencies. The resolution demonstrates both the fragility and the resilience of the open-source ecosystem: fragile because a single corporate decision can create system-wide uncertainty. Resilient because community pressure can reverse that decision quickly when the stakes are sufficiently visible.
6.2 Equifax and consumer AI hardware
The fourth signal is the Equifax breach as security catalyst. The breach covers 145 million consumer records. It was caused by a vulnerability that had a published patch for six months. In scale and nature, this is not a sophisticated attack. It is a failure of basic operational discipline. Patch management, vulnerability scanning, and dependency auditing are not exotic security practices; they are hygiene. The Equifax breach accelerates the DevSecOps movement: the integration of security tooling directly into CI/CD pipelines, dependency scanning into build systems, and vulnerability alerting into developer workflows. In practice, the goal is to make security an automatic consequence of normal development rather than a separate gate.
The fifth signal is the hardware AI specialisation trend beginning to reach consumers. Apple’s Neural Engine in the A11 Bionic is the first indication that AI inference will move to the edge. That means the device in your pocket. Inference will not remain centralised in cloud data centres. The Transformer paper’s algorithmic breakthrough combines with TensorFlow’s Keras integration. Together they make Q3 2017 the beginning of a period in which AI stops being a speciality. AI starts becoming a standard engineering capability.
7. Summary
Q3 2017 is a quarter in which several important technical questions find answers. The container-orchestration question is Kubernetes or something else. It resolves in Kubernetes’s favour, as VMware, Mesosphere, and all major cloud providers commit to it. The serverless maturity question — is Lambda scale a real phenomenon? — receives an empirical answer: more than 10 billion function invocations per month. The React licensing question — will corporate patent clauses constrain open-source adoption? — resolves in favour of permissive licensing under community and competitive pressure.
The open questions that remain at the end of September 2017 are equally significant. Kubernetes is dominant, but operational expertise is scarce. Most teams running Kubernetes in production are still learning how to manage etcd clusters, configure RBAC policies, implement network policies, and operate stateful workloads. The platform’s power is established; the operational maturity of the ecosystem is still developing. Similarly, serverless architectures are maturing technically but challenge the monitoring and debugging tools that operations teams rely on. Functions are ephemeral by design, which makes traditional log-aggregation and APM approaches inadequate. Observability tooling needs to catch up with the deployment model.
7.1 Open questions at the end of September 2017
The Equifax breach leaves a question that will take years to answer. Can the technology industry build security practices that keep pace with attack surface growth? Cloud, microservices, and serverless all create that growth. Each architectural advance reduces operational complexity in one dimension while creating new exposure in another. An organisation running 200 microservices has 200 dependency graphs to audit, 200 container images to patch, and 200 attack surfaces to monitor. The tooling — static analysis, container scanning, runtime security — is emerging, but adoption lags capability.
For practitioners following the AI narrative, Q3 2017 is a moment to pay attention to the Transformer paper. That holds even if language modelling seems distant from current project work. The self-attention mechanism it introduces will reshape not only NLP but vision, recommendation systems, and code generation in the years that follow. The seeds of GPT and BERT are visible in the NIPS 2017 paper submissions circulating in preprint form as the quarter closes. Q3 2017 is the last quarter in which it is reasonable to treat neural language models as a niche research topic.
8. Images
9. Sources
- Kubernetes Blog: “Kubernetes 1.8: Security, Workloads and Feature Depth” — Official release blog post, September 28, 2017.
- React Blog: “React v16.0” — Official React release announcement, September 26, 2017.
- Facebook Code: “Relicensing React, Jest, Flow, and Immutable.js” — Facebook’s MIT relicensing announcement, September 23, 2017.
- TechCrunch: “WordPress to ditch React library over Facebook patent clause risk” — September 15, 2017.
- Kubernetes 1.8 CHANGELOG — Full release notes, September 28, 2017.
- Vaswani et al.: “Attention Is All You Need” — arXiv:1706.03762, June 12, 2017. Introduces the Transformer architecture.
- TensorFlow v1.3.0 Release Notes — Keras integration and XLA improvements, August 17, 2017.
- Apple Newsroom: “Introducing iPhone X” — A11 Bionic chip with Neural Engine, September 12, 2017.
- Enterprise Ethereum Alliance: “EEA Expands to 116 Members” (PDF) — May 19, 2017.
- FTC: “Equifax to Pay $575 Million as Part of Settlement” — FTC press release, July 2019, on the September 2017 Equifax breach (CVE-2017-5638).
- OpenFaaS GitHub Repository — Alex Ellis’s serverless functions framework for Kubernetes and Docker Swarm.
- Serverless Framework GitHub Repository — Multi-cloud serverless deployment framework, v1.19–1.22 during Q3 2017.
- Apollo Client GitHub Repository — Apollo Client 2.0 beta development during Q3 2017; stable release November 2017.
- Andrew Clark: “React Fiber Architecture” — Design document for the Fiber rendering engine introduced in React 16.
- Kelsey Hightower: “Kubernetes the Hard Way” — Step-by-step tutorial for bootstrapping Kubernetes from scratch.
- Alex Ellis Blog — blog.alexellis.io — Author of OpenFaaS; posts on serverless computing, Kubernetes, and functions-as-a-service patterns.
- Martin Fowler: “Microservices” — Foundational article on microservices architecture.
- Wikipedia: React — Licensing section — Historical summary of React’s licensing evolution.
- Wikipedia: Ethereum — History — Covers the Enterprise Ethereum Alliance formation and growth through 2017.
- PyTorch v0.2.0 Release Notes — Introduces torch.distributed for multi-GPU training, August 2017.
- Google Cloud Blog: “Google supercharges machine learning tasks with custom chip” — Cloud TPU announcement and alpha availability.
- Wikipedia: Equifax data breach — Comprehensive summary of the breach disclosed September 7, 2017.
Free ebook
Free AI Video, Generated Locally
Run Wan 2.1 in ComfyUI on your own GPU — the scripts I use, measured times, sample clips. No cloud, no API keys.
No spam. Unsubscribe at any time.
Leave a Reply