Since we already have Etcd as a dependency, I think it makes sense to use etcd to store more of the cluster configuration information. This also means we will have a globally consistent model catalog. For now, basic information like hostname->partition mapping, the number of servers, which models to deploy, etc. can all be pushed into etcd.
This means that I need to change the deploy script to write all of this information to etcd before starting a cluster, and it means that I need to change the bootstrap process to read the config from etcd rather than the local config.yml.
dcrankshaw
commented
9 years ago
Since we already have Etcd as a dependency, I think it makes sense to use etcd to store more of the cluster configuration information. This also means we will have a globally consistent model catalog. For now, basic information like hostname->partition mapping, the number of servers, which models to deploy, etc. can all be pushed into etcd.
This means that I need to change the deploy script to write all of this information to etcd before starting a cluster, and it means that I need to change the bootstrap process to read the config from etcd rather than the local
config.yml.