This week brought a significant architectural change to Flynn: we removed a centralized component responsible for cluster state. Previously every single Flynn host had to start etcd, flannel, and discoverd. Now a single bootstrap phase starts them along with all the other Flynn components.
This not only removes a single point of failure, but allows for faster and more accurate tracking of jobs on the cluster, along with easy updates of these components. This also means the same APIs are now used to manage every single Flynn component. All of these changes will bring improved stability and performance.
Other enhancements included the addition of discoverd reconnects to improve overall resiliency.
- Changed to upstream staticfile buildpack (#1598)
- Refactored host and scheduling to bootstrap once and remove cluster state (#1597, #1602, #1604)
- Added reconnects to discoverd event watches (#1608)
- Fixed broken Ubuntu Virtualbox image (#1600)
- Fixed intermittently failing log following test (#1606)
- Fixed intermittently failing slugbuilder test (#1606)
- Fixed intermittently failing postgres test (#1606)
- Fixed intermittently failing scheduler test (#1606)
- Fixed timeout in postgres test (#1609)