Presentation: Fleet: what comes next for the distributed init system
Fleet, distributed init system developed by CoreOS, has been actively used by production systems. However, many issues still need to be addressed, such as scalability as well as integration with other projects like systemd. This talk will cover the upcoming topics for fleet, to come up with a reasonable plan to continue to work on the project.
Major issues in fleet can be summarized like the below:
* Scalability. Fleet is basically not capable of being scalable well enough to be used with a huge number of units. Most of such issues are resulted from its dependency on etcd v2. A possible solution would be to introduce a gRPC-based communication model. It would be also generally desirable for fleet to be decoupled from etcd as much as possible.
* Reconciliation. This must be one of the critical bottleneck in the fleet agent. Due to a monolithic loop in the fleet agent, units are not able to be propagated right away. Introducing an event-based reconciliation could be one of solutions.
* Systemd units. Although fleet depends heavily on systemd units, there are always subtle mismatches between unit states of fleet and systemd. Limitation of using templates is also one of such issues.