Argo Workflows 2021 Survey Results
The Argo Workflows 2021 Survey ran for three weeks. There were 60 responses. The final NPS score came in at a whopping 66.
Roles
Perhaps unsurprisingly, mostly engineers:
- 32% DevOps Engineer
- 26% Software Engineer
- 15% Architect
- 9% Data Engineer
This very much ties in with what we expected based on the main use cases we know about.
Use Cases
Six use cases dominate:
As respondents lean towards DevOps engineers, this may overstate the proportion of infra automation and CI/CD use cases.
Popular Features
We were pleased to see how popular templates and the API are:
Interestingly, I think we expected more people to be using one of the different language clients. Both semaphore and mutex we’re lower than expected, but these are quite new features.
Scale
- While most users are running 10s or 100s of workflows a day, 3 users report running 1000s per day.
- The majority of users are running 10s or 100s of pod per workflow, but 2 users reported running over 10,0000 per workflow.
Barriers To Adoption
YAML was the main challenge; many users are more familiar with Python (7 mentions).
Getting familiar with cloud-native and containers was also highlighted as a challenge (3 mentions).
“Convincing folks that containers aren’t scary and YAML isn’t evil”
Additions and Improvements
Provide a Python DSL (5 mentions). It’s worth noting that a Python DSL for Argo Workflow already exists in the form of Couler (learn about authoring workflows in Python).
“Honestly, something like a good python DSL would be fantastic”
Improved UI (3 mentions). Argo Workflows v3.0 has a revamped user interface.
Documentation was mentioned as an area for improvement as well (3 mentions).
“I do find that often features we want exist already but can be a little difficult to discover”
“I think the docs sometimes feel a little bare or like they don’t have enough structure.”
“Improve UI, multi-cluster, docs”
Why Argo?
Users love that it is Cloud-native/Kubernetes (28 mentions)
The best workflow manager (6-ish mentions)
“Best workflow manager on K8S”
“the best K8s orchestrator we found”
“one of the best cloud native tools i’ve ever seen”
“better than other frameworks that were evaluated”
“Best workflow manager on K8s.”
“Because its the best to manage DAGs in the industry”
Comparatively lightweight and easy to use (5 mentions)
“…lightweight, kubernetes native. Great community.”
“we wanted more flexibility...”
“Its way faster…”
“so much more powerful…”
Works well with Argo CD (5 mentions).
“We picked it for the easy integration with ArgoCD”
“We were already using ArgoCD. The doc/community/concepts felt good. We managed to get satisfying state quickly.”
Next Steps
We’re currently considering:
- Who wants a Python DSL? What would they use it for? We already have Couler — do we need to promote this solution more? Do we need to provide 1st-class support for Python?
- What areas of the documentation need improving? How can we determine this? Can the community help?
