Results from the Cloud Native Computing Foundation’s (CNCF) latest survey, conducting in September/October 2019, are worthy of further discussion. Hopefully, our experience digging into previous editions of the study provides a perspective you won’t find anywhere else. The 1,337 respondents from the CNCF community are by their very nature more likely than other IT professionals to have advanced cloud native practices.
Here are few quick takes:
- Adoption of containers in production rose to from 73% in 2018 to 84% in 2019. Among this group, those running at least 250 containers rose from 46% in 2018 to 58% in 2019. From 2017 to 2019, the number of respondents with more than 50 machines (physical or virtual) in their fleet rose from 77% in 2017 to 81% in 2019.
- Implication: Container adoption appears to have mitigated the growth of VMs that need to be managed. However, be wary of claims that the raw number of machines being managed will decline.
- Production-level use of hosted serverless platforms rose from a little over to 30% to 40% year-over-year. However, among these users, adoption of AWS Lambda (70% to 53%), Google Cloud Functions (25% to 18%), and Azure Functions (20% to 14%) all dropped.
- Use of installable serverless platforms rose 6% in 2018 to 10% in 2019. Among these users, Knative is the top choice.
- There has been some degree of non-response to this question, which means that the reported adoption rates are likely a few percentage points lower than if responses were required.
- Implication: Initially companies evaluated multiple FaaS offerings at the same time. Opportunities to increase the breadth of serverless adoption in multicloud environments may be limited to unique use cases and non-compute related services.
- When presented with a list of eight service mesh-related products or projects, 18% said they are using at least one of them in production. Forty-seven percent are also evaluating one of these technologies. About 20% of the study did not respond to the question, so the actual adoption rate is most likely lower than reported in the report’s charts. The overall level of use and consideration appears to be within the range of what we have previously written about.
- Over 40% of the people that answered this question are using HashiCorp’s Consul. About 27% mentioned using Istio in production, with Netflix OSS Zuul and Linkerd rounding out the top four.
- About 60% of the study said they use ngnix as a Kubernetes ingress provider. However, it is possible that many respondents cited the technology’s use even if it is primarily deployed to environments not related to Kubernetes. HAproxy and F5 Networks saw drops in adoption while Envoy and Traefik use increased.
- The survey also about the uptake of CNCF’s graduated, incubating and sandbox projects. Unfortunately, the way the data is presented in the report we can’t make an apples-to-apples comparison of how generic Envoy and Linkerd adoption is being adopted in comparison to more specific service mesh applications.
- Implication: Many Consul users were likely using it for service discovery even before they had heard of the service mesh term. HashiCorp’s marketing and product development efforts appear to be capitalizing as they pitch it as a multicloud, service-mesh technology. The study’s raw data probably provides rare insights into how and when Istio, Envoy and other tools are being used in the same stack. The New Stack readers should stay tuned for more analysis when that happens.
- The report concludes that “an increase in both the number and reliability of CI/CD tools is driving a decrease in manual release cycles and an acceleration of release cycles.”
- A cursory reading of one chart shows that respondents with daily or weekly release cycles rose from 35% in 2018 to 55% in 2019. However, much of rise can be ascribed to a large decline of the percentage of people who didn’t answer the question last year or otherwise were coded as “N/A” in the open-ended version of the question in 2018. People that said their release cycle is “ad hoc” likely rose partly because the response was available in the multiple-choice question.
- Another chart (not included in this post) indicates that people with manually managed release cycles dropped from 27% to 14%. However, the drop is most likely because this year’s question specifically offered hybrid (both manual and automated) as an answer choice.
- Implication: Developer and deployment velocity is hard to measure. The latest CNCF report does a better job at capturing the nuance that software releases can often be done on an ad hoc basis. Many CI/CD systems are not 100% automated, so there should not be a binary automated versus not automated category.