The big news from this past week is that we shipped a complete rewrite of the scheduler. While this will not have any noticeable immediate impact beyond clearing up a few lingering bugs, it includes a number of enhancements under the hood that will enable future features.
The dashboard’s app history view now supports rolling back to previously deployed releases.
We also continued our focus on cleaning up bugs, with several more bugfixes shipped.
We shipped a long-awaited feature this week: the ability to run multi-node Flynn clusters on one machine, e.g. on a local Vagrant instance. This means that developers can test multi-node cluster functionality locally, rather than having to spin up multiple cloud instances.
We also shipped a number of browser compatibility fixes and usability enhancements as we work towards a polished 1.0 release.
This week brought two significant changes to Flynn.
The component that powers service discovery in Flynn, discoverd, now includes
Raft in-process. This was the final component that required etcd, which had been
a source of instability in Flynn clusters for some time. Etcd is no longer
shipped with Flynn. This change had increased cluster stability substantially.
This is the first of several expected changes to our use of consensus
algorithms. In the future we will only depend on Raft for leader election and
consistent service state and instead use gossip for discovering service
instances (which do not require consistency).
The installer can now install on arbitrary hosts via SSH. Until now the
installer could only target public clouds like AWS, Azure, and Digital Ocean.
Now you can install Flynn on any machine to which you have access by running