Load and Performance Testing
Juju should be proven to scale to environments of at least 10,000 machines.
Juju is a solution for managing the deployment of Openstack clouds, which can sprawl into tens of thousands of servers. Juju should comfortably handle these sized environments.
Blueprint information
- Status:
- Complete
- Approver:
- Mark Ramm
- Priority:
- Undefined
- Drafter:
- Dave Cheney
- Direction:
- Needs approval
- Assignee:
- None
- Definition:
- Obsolete
- Series goal:
- None
- Implementation:
-
Unknown
- Milestone target:
- None
- Started by
- Completed by
- Katherine Cox-Buday
Whiteboard
Commercial Juju customers want to be able to use Juju to manage their MaaS based Openstack deployments. These environments could exceed 10k physical machines. Juju must be capable of operating within expectations with environments of at least 10k active agents across a mixture of charm deployments (Hadoop, Openstack, etc).
Previous load testing shows a hard limit of 1200 machines (~2400 agents/mongodb connections), with the limit of acceptable performance and responsiveness of the cli tools much lower.
However, we are now able to reasonably scale into 8,000 agents (4,000 machine + 4,000 user agents), though some aspects start to slow down.
[ASSUMPTIONS]
*
[RISKS]
* It may be difficult to acquire the resources to validate a 10k machine deployments
[OUT OF SCOPE]
* Scaling higher than 10k machines is out of scope
[USER ACCEPTANCE]
* Load testing similar with 0.2k, 2k, 5k, and 10k, hadoop tests
* Develop a tool to load test the api directly without launching machines. [done]
[RELEASE NOTE/BLOG]
* Blog posts, possibly co branded.
