Back in July, we opened and shared the 2022 Serverless Community Survey. Our goal was to better understand how serverless is currently used by teams all over the industry. The survey is now closed. Let's dive into the results!
Context
We make the Serverless Framework. Since 2015, our definition of "serverless" is building applications on managed services that scale to zero to radically reduce overhead in crafting software.
Due to the design of Serverless Framework, that definition is interpreted in our community as using lots of AWS Lambda and other managed services, mostly on AWS, to build APIs, Data Processing Jobs, Event-Driven Architectures, Websites, and more. Many Serverless Framework community members filled in this survey. As a result, this survey is weighted toward AWS, AWS Lambda, and Serverless Framework, and represents trends on how development teams use them.
To broaden our outreach, this survey was shared on Twitter, Reddit communities (serverless, AWS, Azure, GCP), Slack (Serverless Framework, AWS user groups, AWS CDK), serverless newsletters, Hacker News, and shared by serverless community members as well.
Overall, the survey collected over 300 responses, evenly distributed in terms of company size (only 5% of respondents were solo developers). 90% of respondents declared using serverless in production, 9% experimenting with it, and only 3% not using serverless at all. As such, most of our analysis focuses on "in-production serverless usage".
Cloud Providers
TweetGiven that our survey results are partial to AWS, we took the opportunity to look at cloud providers used with AWS. 20% of respondents use AWS with another cloud provider, and 11% with two other cloud providers.
Serverless in production
Since 90% of respondents use serverless in production, we can dig into their serverless usage. 56% of them have been running serverless in production for over two years. Organizations "new to serverless," i.e., using serverless in production for less than six months, only account for 5%. All that confirms serverless has matured and is on its way to becoming a boring technology choice.
.png)
Serverless has also gone from the "adoption" phase to the "scaling up" phase: more than half of serverless teams have over 50 serverless functions in production. 39% have more than 100 functions.
The trend is even more drastic when looking at "how many serverless services companies run" compared to the organization size:
- Solo developer: 5 services
- 2 to 10 employees: 7 services
- 11 to 50 employees: 35 services
- 51 to 500 employees: 33 services
- 500+ employees: 41 services
Note: by "service" we mean a collection of functions deployed together. This can be a standalone application or a single service from a micro-services architecture.
Serverless challenges
When asked to rank the biggest challenges working with serverless, we found three clear winners. Interestingly, we find the same top 3 challenges when filtering for experienced serverless builders or "new to serverless" respondents.
We also believe that serverless monitoring and debugging should be simpler. This is why we are working on Serverless Console, a next generation solution for cloud development and observability.
Serverless tooling
AWS CloudWatch is the most widely used tool for monitoring and debugging, with 86% adoption across respondents running serverless in production.
When compiling the numbers another way, we find out that 28% do not use anything for monitoring other than AWS CloudWatch or X-Ray.
As for deploying serverless applications, Serverless Framework comes up as the most popular tool, followed by CloudFormation and AWS CDK:
- Serverless Framework: 67%
- CloudFormation: 51%
- AWS CDK: 36%
- Terraform: 23%
- AWS SAM: 19%
- SST: 14%
- Pulumi: 5%
- Other: 12%
An obvious thing to note is that results might be skewed towards Serverless Framework since… we ran the survey. That being said, Datadog's 2021 State of Serverless reported that Serverless Framework is used by 90% of serverless organizations using CloudFormation. So, while we will refrain from tooting our own horn, the numbers above might not be so far off.
If you are wondering, the numbers add up to more than 100% because organizations might use more than one tool. CloudFormation is also used with Serverless Framework and other tools, explaining its high usage.
Languages and runtimes
When looking at which languages are the most popular with serverless, JavaScript and TypeScript are the clear winners.
However, the overlap between languages is interesting: the sum of all percentages exceeds 200%. Can we infer that serverless increases the chances of mixing multiple languages in one project? Or maybe that serverless adoption at larger companies is first driven by JavaScript/TypeScript teams and then spreads to other teams?
When excluding smaller companies (< 50 employees) from the analysis, JavaScript drops to 55% and .Net shots up to 12%
When splitting the results by IaC tools, results vary slightly as well:
- Serverless Framework has more JavaScript users (73%) than TypeScript (67%)
- AWS CDK has more TypeScript users (85%) than JavaScript (69%)
- Terraform's 2nd most popular language is Python (55%) instead of TypeScript
Container image functions
We looked at Serverless Framework usage data for additional insights. AWS announced container image support for AWS Lambda in 2020. Two years later, only 2% of Serverless Framework deployments involve container images.
Use-cases
.png)
We asked respondents to select which use-cases they are building, and which they are building with serverless. That lets us get a glimpse at which use-cases are the most common and which are solved the most with serverless.
There is a lot to unpack, so let's start from the top: the most popular use-cases. HTTP API, queues (asynchronous tasks), scheduled tasks (cron), and event bus (messaging) are implemented by more than 80% of respondents. Additionally, these popular use-cases are implemented with serverless in 85% of cases.
The least popular use-cases overall are at the bottom: edge computing, machine learning, and internet of things, implemented by less than 30% of respondents. Machine learning is also the use-case with the lowest serverless adoption (45%).
More generally, we can identify the following learnings:
- Serverless teams implement many different use-cases, not just APIs.
- Serverless teams tend to apply serverless everywhere.
- Some use-cases, like frontend websites, ETL, and machine learning, are less likely to be implemented with serverless (71%, 67%, 45% respectively).
Databases
While DynamoDB is the most common database used in production serverless applications, relational databases (MySQL, PostgreSQL…) are used by 44% of respondents.
Note: databases such as Azure Cosmos DB, FaunaDB, and Cassandra were not included in the illustration above because their reported usage was less than 2%.
We also looked at Serverless Framework usage metrics for additional data points. We found that relational databases are used as much as DynamoDB in Serverless Framework projects in production.
Application Architecture and Development
Micro-functions (single-purpose functions) vs mono-lambdas (multi-purpose functions) is a debate as old as serverless. While there is a clear majority of developers going for the single-purpose functions, adoption of that pattern is not overwhelming, and some teams like to mix and match.
When looking at applications composed of multiple services, most organizations go with the monorepository approach:
- Monorepository containing multiple services: 58%
- Multiple repositories (one per service) deployed independently: 36%
- Multiple repositories (one per service) with deployment dependencies: 19%
The bigger the company, the more people work on the same service. However, on average less than eight people work on the same service. That number drops to four people for companies with less than 50 employees.
Finally, development practices are also a popular topic of debate. It turns out there is not one way to develop serverless applications. Results are almost evenly split between developing locally or in the cloud, even inside teams.
Try Serverless Console
Real-time dev mode provides streaming logs from your AWS Lambda Functions.
