Dmytro Sen created AMBARI-10316:
-----------------------------------
Summary: Stack-advisor mechanism when recommending
configuration-dependencies has incorrect services.json content
Key: AMBARI-10316
URL: https://issues.apache.org/jira/browse/AMBARI-10316
Project: Ambari
Issue Type: Bug
Components: ambari-server, stacks
Affects Versions: 2.1.0
Reporter: Dmytro Sen
Assignee: Dmytro Sen
Priority: Critical
Fix For: 2.1.0
When the POST call is made for recommending {{configuration-dependencies}}, we
generate the services.json file with following extra content:
{code}
"changed-configurations" : [ {
"type" : "yarn-site",
"name" : "yarn.scheduler.maximum-allocation-mb"
} ],
"depended-configurations" : [ {
"type" : "mapred-site",
"name" : "yarn.app.mapreduce.am.resource.mb"
}, {
"type" : "yarn-site",
"name" : "yarn.scheduler.maximum-allocation-mb"
}, {
"type" : "mapred-site",
"name" : "yarn.app.mapreduce.am.admin-command-opts"
}, {
"type" : "mapred-site",
"name" : "mapreduce.map.memory.mb"
}, {
"type" : "mapred-site",
"name" : "yarn.app.mapreduce.am.command-opts"
}, {
"type" : "mapred-site",
"name" : "mapreduce.reduce.java.opts"
}, {
"type" : "mapred-site",
"name" : "mapreduce.reduce.memory.mb"
}, {
"type" : "mapred-site",
"name" : "mapreduce.task.io.sort.mb"
}, {
"type" : "mapred-site",
"name" : "mapreduce.map.java.opts"
} ]
{code}
The configuration depended-by should not be added on to the services.json like
this. All the configuration dependencies should be got from manipulating the
{{org.apache.ambari.server.api.services.stackadvisor.commands.StackAdvisorCommand.GET_SERVICES_INFO_URI}}
value. Changing the value to include configuration depends_on and depended_by
information. Then the python script should build the dependency-chain inside
itself instead of doing it in Java and adding to services.json.
The reason this is important is that for future functionality, we need
services.json and hosts.json to faithfully represent the cluster's services and
hosts information. The custom stack-advisor scripts should have logic to
interpret it for their own needs. If we do some processing in Java, then
stack-providers will not have chance to make their own interpretations.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
