7. Cluster Manifest:
Host List
• List of hosts
– Can be fully specified
– Or, can be a set of requirements
– Or, can even be non-existent
{
"schemaVersion:" : "1", …
"context" : […],
"hostGroups" : [
{
"name" : "masterHosts",
"members" : {
"count" : "1",
"hosts" : [
{ "FQDN" : "host1.domain1.com", "ip" : "" }
]
},
"properties" : […]
},
{
"name" : ”slaveHosts",
"members" : {…},
"properties" : […]
},
{
"name" : "clientHosts",
"members" : {…},
"properties" : [
{ "name" : "host_type", "value" : "High-CPU Medium" }
]
},
...
]
}
8. Cluster Manifest:
Host Component Mapping
• A mapping of components to hosts
– Simple component mapping to named hosts
– A set of constraints that can be used to find best
match (e.g. evaluate against host properties)
• System resources
– users, groups, ports, etc.
• Host specific configuration
– Non-homogeneous cluster
10. Scenarios
• Define cluster templates
– and, host specific templates
• On demand cluster creation
– Cluster extension (e.g. add Datanodes)
• Export cluster manifest
• A uniform “language” across cluster managers
and environments
11. Cluster Blueprint
• Blueprint is manifest with “holes”
– Typically
• Hostnames
• Config parameters that use hostname
– But, any config params that a Hadoop admin
deems necessary to be parameterized
• Blueprint = Manifest + Parameter Values
13. What’s Next?
• Apache Ambari JIRA 1783, is tracking this
project
– https://issues.apache.org/jira/browse/AMBARI-
1783
– Comments and suggestions, welcome
• In next releases, we will enhance Ambari to
add support for manifest and blueprints