O’Reilly online learning comprises details about the developments, matters, and points tech leaders want to look at and discover. It’s additionally the information supply for our annual utilization research, which examines the most-used matters and the highest search phrases.[1]
This mixture of utilization and search affords a contextual view that encompasses not solely the instruments, strategies, and applied sciences that members are actively utilizing, but in addition the areas they’re gathering details about.
Present indicators from utilization on the O’Reilly on-line studying platform reveal:
- Python is preeminent. It’s the one hottest programming language on O’Reilly, and it accounts for 10% of all utilization. This yr’s progress in Python utilization was buoyed by its growing recognition amongst knowledge scientists and machine studying (ML) and synthetic intelligence (AI) engineers.
- Software program structure, infrastructure, and operations are every altering quickly. The shift to cloud native design is reworking each software program structure and infrastructure and operations. Additionally: infrastructure and operations is trending up, whereas DevOps is trending down. Coincidence? In all probability not, however solely time will inform.
- ML + AI are up, however passions have cooled. Up till 2017, the ML+AI subject had been amongst the quickest rising matters on the platform. Development continues to be sturdy for such a big subject, however utilization slowed in 2018 (+13%) and cooled considerably in 2019, rising by simply 7%. Inside the knowledge subject, nonetheless, ML+AI has gone from 22% of all utilization to 26%.
- Nonetheless cloud-y, however with a chance of migration. Robust utilization in cloud platforms (+16%) accounted for many cloud-specific progress. However sustained curiosity in cloud migrations—utilization was up nearly 10% in 2019, on prime of 30% in 2018—will get at one other essential rising development.
- Safety is surging. Mixture safety utilization spiked 26% final yr, pushed by elevated utilization for 2 safety certifications: CompTIA Safety (+50%) and CompTIA CySA+ (+59%). There’s loads of safety dangers for enterprise executives, sysadmins, DBAs, builders, and so forth., to be cautious of.






In programming, Python is preeminent
In 2019, as in 2018, Python was the most well-liked language on O’Reilly on-line studying. Python-related utilization grew at a strong 6% tempo in 2019, a slight drop from 2018 (+10%). After a number of years of regular climbing—and after outstripping Java in 2017—Python-related interactions now comprise nearly 10% of all utilization.
However Python is a particular case: this yr, greater than in yr’s previous, its progress was buoyed by curiosity in ML. Utilization particular to Python as a programming language grew by simply 4% in 2019; against this, utilization that needed to do with Python and ML—be it within the context of AI, deep studying, and pure language processing, or together with any of a number of widespread ML/AI frameworks—grew by 9%. The laggard use case was Python-based net improvement frameworks, which grew by simply 3% in utilization, yr over yr.






That is per what we’ve observed elsewhere: Python has acquired new relevance amid sturdy curiosity in AI and ML. Together with R, Python is without doubt one of the most-used languages for knowledge evaluation. From pre-built libraries for linear or logistic regressions, resolution timber, naïve Bayes, k-means, gradient-boosting, and so forth., there’s a Python library for nearly something a developer or knowledge scientist may have to do. (Python libraries aren’t any much less helpful for manipulating or engineering knowledge, too.)
Apparently, R itself continues to decline. R-related utilization on O’Reilly on-line studying fell by 8% between 2017-18 and by 6%, year-over-year, in 2019. It’s possible that R—very similar to Scala (-33% in utilization in 2018-19; -19% in utilization 2017-18)—is a casualty of Python. True, it may appear troublesome to reconcile R’s decline with sturdy curiosity in AI and ML, however think about two elements: first, ML and statistics aren’t the identical factor, and, second, R isn’t, primarily, a developer-oriented language. R was designed to be used in educational, scientific, and, extra lately, industrial use circumstances. As statistics and associated strategies grow to be extra essential in software program improvement, extra programmers are encountering stats in programming courses. On this context, they’re extra possible to make use of Python than R.
Curiosity in some languages appears to be trending up, and curiosity in others, down. Exhibit A: Java-related utilization dropped by a noteworthy 13% between 2018 and 2019. Is that this the harbinger of a development? Not essentially: Java-related searches elevated by 5% between 2017 and 2018. However, Java’s cousin, JavaScript, additionally seems to be in decline. True, theirs is just a conceptual relation, however curiosity in JavaScript, too, really does seem to be waning: JS-related utilization dropped on O’Reilly on-line studying by 4% between 2017-2018 and by 7% between 2018-19. It’s doable that microservices structure is hastening the transfer to different languages (comparable to Go, Rust, and Python) for net properties.
Among the many JavaScript-based net software frameworks, React gained in recognition (+4% in utilization) as Angular (-12% in utilization) slipped between 2018 and 2019. Vue.js—a competitor to each React and Angular—settled all the way down to regular progress (+8% in utilization) in 2018-19, after nearly doubling in utilization (+97%) between 2017-18.
One doable trend-in-the-making is that of a slowing Go, which—following a number of years of speedy progress in utilization (together with +14% from 2017 to 2018)—cooled down final yr, with utilization rising by a mere 2%. However Go is now the sixth most-used programming language, trailing solely Python, Java, .NET, and C++. Drop .NET from the tally on methodological grounds[2], and Go cracks the highest 5.
Traits in software program structure, infrastructure, and operations
Cloud native design is a brand new mind-set about software program and structure. However the shift to cloud native has implications not just for software architecture, however for infrastructure and operations, too. It exploits new design patterns (microservices) and adapts present strategies (service orchestration) with the objective of attaining cloud-like elasticity and resilience in all environments, cloud or on-premises. O’Reilly Radar makes use of the time period “Next Architecture” to explain this shift.
It’s in opposition to this backdrop that what’s taking place in each software program structure and infrastructure and ops should be understood. Within the generic software program structure subject, utilization within the containers subject elevated in our 2019 evaluation, rising by 17%. This was only a fraction of its 2018 progress charge (+56% in utilization), however spectacular nonetheless. Kubernetes has emerged because the de facto resolution for orchestrating providers and microservices in cloud native design patterns. Utilization in Kubernetes surged by 211% in 2018—and grew at a 40% clip in 2019. Kubernetes’ mother or father subject, container orchestrators, additionally posted sturdy utilization progress: 151% in 2018, 36% this yr—nearly all on account of curiosity in Kubernetes itself.



This additionally helps clarify elevated utilization within the microservices subject, which grew at a 22% clip in 2019. True, you don’t essentially want microservices to “do” cloud native design; at this level, nonetheless, it’s troublesome to disentangle the 2. Most cloud native design patterns contain microservices.
These developments are additionally implicated within the rise of infrastructure and ops, which displays each the constraints of DevOps and the challenges posed by the shift to cloud native design. Infrastructure and ops utilization was the quickest rising sub-topic beneath the generic programs administration subject. Surging curiosity in infrastructure and ops additionally explains declining utilization within the configuration administration (CM) and DevOps subject areas. The preferred CM instruments are DevOps centered, and, like DevOps itself, they’re declining: utilization within the CM subject dropped considerably (-18%) in 2019, as did nearly all CM instruments. Ansible was least affected (-4% in utilization), however Jenkins, Puppet, Chef, and Salt every dropped off by 25% or extra in utilization. It could’t be a coincidence that DevOps utilization declined once more (-5%) in 2019, following a 20% decline in 2018.



The emergence of infrastructure and ops means that organizations may be having bother scaling DevOps. DevOps goals to supply programmers who can work competently in every of the layers in a system “stack.” In apply, nonetheless, builders are typically much less dedicated to DevOps’ operations part, a indisputable fact that gave delivery to the thought of site reliability engineering (SRE). Even when the “full stack” developer isn’t a unicorn, she actually isn’t commonplace. Organizations see infrastructure and ops as a practical, ops-focused complement that picks up exactly the place DevOps tends to fail.
A drill-down into knowledge, AI, and ML matters
The outcomes for data-related matters are each predictable and—there’s no different solution to put it—complicated. Beginning with knowledge engineering, the spine of all knowledge work (the class contains titles masking knowledge administration, i.e., relational databases, Spark, Hadoop, SQL, NoSQL, and so forth.). In combination, knowledge engineering utilization declined 8% in 2019. This follows a 3% drop in 2018. Each years had been pushed by declining utilization of knowledge administration titles.
After we look extra particularly at knowledge engineering matters, excluding knowledge administration, we see a small share, however strong progress in utilization, up 7% in 2018 and 15% in 2019 (see Determine 7).
Inside the broad “knowledge” subject, knowledge engineering (together with knowledge administration) continues as the subject with essentially the most share, garnering about one-twelfth of all utilization on the platform. That is nearly double the utilization share of the information science subject, which recorded an uptick in utilization (+5%) in 2019, following a decline (-2%) in 2018.
Elsewhere, curiosity in ML and AI retains rising, albeit at a diminished charge. To wit: the mixed ML/AI subject was up 7% in utilization in 2019, about half its progress (+13%) in 2018.



Sarcastically, the energy of ML/AI may be much less evident in data-specific matters than in different subject areas, comparable to programming languages, the place rising Python utilization is—to a big diploma—being pushed by that language’s usefulness for and applicability to ML. However ML/AI-related matters comparable to pure language processing (NLP, +22% in 2019) and neural networks (+17%) recorded sturdy progress in utilization, too.
Information engineering as a activity actually isn’t in decline. Curiosity in knowledge engineering in all probability isn’t declining, both. If something, knowledge engineering as a apply space is being subsumed by each knowledge science and ML/AI[3]. We all know from other research that knowledge scientists, ML and AI engineers, and so forth., spend an outsized proportion of their time discovering, getting ready, and engineering knowledge for his or her work. We’ve seen that widespread instruments and frameworks often incorporate knowledge engineering capabilities, both within the type of automated/guided self-service options or (within the case of Jupyter and different notebooks) a capability to construct and orchestrate knowledge engineering pipelines that invoke Python, R (by way of Python), and so forth., libraries to run knowledge engineering jobs concurrently or, if doable, in parallel.
Phrases that correspond with old-school knowledge engineering—e.g., “relational database,” “Oracle database options,” “Hive,” “database administration,” “knowledge fashions,” “Spark”—declined in utilization, year-over-year, in 2019. A few of this decline was a operate of bigger, market-driven elements. We all know from our analysis that Hadoop and its ecosystem of associated tasks (comparable to Hive) are in the midst of a protracted, years-long decline. This decline is borne out in our utilization numbers: Hadoop (-34%), Hive (additionally -34%), and even Spark (-21%) had been all down, considerably, year-over-year.
We focus on possible causes for this decline in additional element in our analysis of O’Reilly Strata Conference speaker proposals.
Cloud persevering with to climb
Curiosity in cloud-related ideas and phrases continues to extend on O’Reilly on-line studying, albeit at a slower charge. Cloud-related utilization surged by 35% between 2017 and 2018; it grew at lower than half that charge (17%) between 2018 and 2019. This slowdown means that cloud as a class has achieved such a big share that (mathematically) any further progress should happen at a slower charge. In cloud’s case, whereas progress is slower, it’s nonetheless sturdy.



Curiosity in cloud service supplier platforms mirrors that of the trade as an entire: Amazon and AWS-related utilization elevated by 14%, year-over-year; Azure utilization, however, grew at a speedier 29% clip, whereas Google Compute Platform (GCP) surged by 39%. Amazon controls rather less than half (per Gartner’s 2018 numbers) of the general marketplace for cloud infrastructure-as-a-service (IaaS) choices. It, too, has reached the purpose at which speedy progress turns into mathematically prohibitive. Each Azure and GCP are rising a lot quicker than AWS, however they’re additionally a lot smaller: Azure notched practically 61% progress in 2018 (per Gartner), good for greater than 15% of the IaaS market; GCP, at round 60% progress, accounts for 4% of IaaS share.
Additionally intriguing: cloud-specific curiosity in microservices and Kubernetes grew considerably final yr on O’Reilly. Microservices-related utilization was up 22%, yr over yr, following a decline in 2018. Kubernetes utilization was up by 38%, yr over yr, following a interval of explosive progress (+190%) from 2017 to 2018. Each developments mirror what we’re seeing by way of consumer surveys and different research efforts: specifically, that microservices has emerged as an essential part of cloud native design and improvement.
The larger takeaway is that the important tendency of contemporary software program structure—specifically, the precedence it provides to free coupling in emphasizing abstraction, isolation, and atomicity—is eliding the boundaries between what we consider as “cloud” versus “on-premises” contexts. We see this by way of sustained curiosity in microservices and Kubernetes in each on-premises and cloud deployments.
That is the logic of cloud native design: particular deployment contexts will nonetheless matter, in fact—which options or constraints do builders have to take note of once they’re growing for AWS? For Azure? For GCP? However the clear boundaries that used to demarcate the general public cloud from the non-public cloud are beginning to disappear, simply as those who distinguish on-premises non-public clouds from standard on-premises programs are falling away, too.
Surging curiosity in safety
Safety utilization (+26%) grew considerably in 2019 (see Determine 2). A few of this was pushed by elevated utilization within the CompTIA Safety+ (50%) and CompTIA Cyber Safety Analyst (CySA+, 59%) matters.
Safety+ is an entry-level safety certification, so its progress might be attributed to elevated utilization by sysadmins, DBAs, software program builders, and different non-specialists. Whether or not it’s to flesh out their full-stack bona fides, deal with new job (or regulatory) necessities, or just to make themselves extra marketable, Safety+ is a reasonably easy certification course of: go the examination and also you’re licensed. CySA+, however, is comparatively new. This may clarify the explosion of CySA+-related utilization in 2018 (+128%), in addition to final yr’s sturdy progress. In contrast to the CISSP and other popular certifications, CySA+ recommends, however doesn’t require, real-world expertise. Like Safety+, it’s one other certification sys admins, DBAs, builders, and others can decide as much as burnish their bona fides.
Certifications weren’t the one factor driving security-related utilization on O’Reilly in 2019. A rash of vulnerabilities and potential exploits, too, had some influence. If 2018 (+5% progress in safety utilization; +22% progress in search) gave us Meltdown and Spectre, 2019 gave us sobering details about the far-reaching implications of Meltdown and, especially, Spectre. For 2019, security-specific utilization (+26%) and search (+25%) elevated accordingly. System and database directors, CSAs, CISSPs, and others had been eager to amass knowledgeable, detailed data particular to patching and hardening their weak programs to guard in opposition to no less than 13 different Spectre and 14 different Meltdown variants, in addition to to mitigate the potentially huge performance impacts associated with these patches. Builders and software program architects had questions on rewriting, refactoring, or optimizing their code to deal with these similar considerations. In opposition to this backdrop, the spike in security-related utilization is sensible.
There was an amazing deal happening with respect to data safety and knowledge privateness, too. In any case, not solely was 2019 the primary full yr for which the EU’s omnibus GDPR regime was binding, however—as of January 1, 2019—updates to Canada’s GDPR-like PIPEDA regime formally kicked in, too. The sweeping California Consumer Privacy Act (CCPA), which has been known as California’s GDPR, went into impact on January 1, 2020.
Taken collectively, an evaluation of those developments appears to assist a glass-half-full evaluation of the state of safety as we speak. If the sustained progress in safety utilization on O’Reilly is a dependable indicator, it’s doable that safety could, lastly, be getting the eye it deserves in an more and more digital world. It’s doable that organizations have accepted that the monetary and reputational penalties entailed by a knowledge breach or high-profile hack are simply too pricey to threat, and that cash spent on data safety is, on steadiness, cash properly spent.
The identical evaluation additionally lends itself to a glass-half-empty evaluation, nonetheless: specifically, that safety spending is cyclical; {that a} confluence of circumstances has helped to spice up safety spending; and that—let’s be trustworthy—organizations tend to bounce back from high-profile security incidents. Solely time (or future installments of this survey) will inform.
Concluding ideas
It’s exhausting to think about that the most well liked developments of 2019 gained’t be reprising their roles, in kind of the identical pecking order, in subsequent yr’s evaluation. Programming languages come into and exit of vogue, however Python seems poised to continue to grow at a gentle charge as a result of it’s without delay protean, adaptable, and straightforward to make use of. We see this within the widespread use of Python in ML and AI, the place it has supplanted R because the lingua franca of knowledge engineering and evaluation.
The identical is true of ML and AI. Even when (as some naysayers warn) the following AI winter is nigh upon us, it’s exhausting to think about curiosity in ML and AI really fizzling out anytime quickly. The identical might be mentioned about developments in software program structure and, particularly, infrastructure and operations. They’re every websites of ceaseless innovation. Their practitioners might be hard-pressed to maintain up with what’s taking place.
It’s useful to think about what’s sizzling and what’s not by way of a modified “Overton Window.” The Overton Window circumscribes the human cognitive bandwidth that’s obtainable in a sure place at a sure time. No mixture of insurance policies—or points, or developments—can exceed greater than 100% of obtainable bandwidth. That is true, to a level, of the exercise on O’Reilly on-line studying, too. A decline in utilization doesn’t need to correlate with a decline in use (or usefulness) in apply. It’s simply being crowded out by different, emergent developments.
This additionally underscores why the decline in a bellwether subject comparable to JavaScript may be massively important. Even when these matters are not websites of speedy and sustained innovation, they’re likewise essential for day-to-day use circumstances, particularly with respect to general-purpose information-gathering or extra specialised downside fixing. It isn’t that JavaScript is much less essential than it was once; in any case, React, Angular, and Vue.js are websites of improvement and innovation, and all three are primarily based on JavaScript. It’s that we’re coming to know and relate to JavaScript—or knowledge engineering, or Docker, or DevOps and alter administration—another way. We’re appropriating them in a different way.
It’s this distinction that Radar goals to seize. Not the change that’s apparent for everybody to see, however the coalescence of change itself because it’s taking place.