Data locality and High availability is possible. The tan boundary represents a Node Group; information about Node Groups would be contained in ETCD and they would be controlled with an operator in the Data DSL.
Data on each node (blue) could be simultaneously updated via update stream to each daemon local database running on a node. Pods would have affinity for nodes.
On container failure - container would come back in the same pod.
On pod failure - pod would come back on the same node.
On node failure - the remaining nodes in the group would pick up the remaining tasks while the Node was re-provisioned.
Many node groups could be used to shard up a data set; it would not need to be the whole data set for each blue node.
No comments:
Post a Comment