Currently, zero-downtime upgrade is highly demand feature from OpenStack users. Current implementation status in OpenStack is variable between projects and almost projects have not fully implemented rolling upgrade. One among Kolla's deliverables is OpenStack container-based deployment tool, thus adding zero-downtime upgrade support in Kolla is reasonable. Our approach is to implement in Kolla natively without needing support from individual OpenStack service. We propose 2 solutions for this problem. First approach achieve real "zero-downtime" from user PoV if upgrade time is short enough, but fail if upgrade time is too long, and it consumes much memory if system is under heavy load during upgrading. Second approach requires more resources than the first one and user will see a short downtime but does not suffer timeout problem. So we can combine two solutions to get advantages from each. The spec and demo for the 1st solution has been proposed and in review queue of Kolla community.
- Rolling upgrade implemenation status in OpenStack eco-system.
- Two approaches for achieving zero-downtime upgrade in Kolla (with container technologies).
- Demo for one approach.