Erik Bergenholtz created AMBARI-11209:
-----------------------------------------
Summary: Blueprints processor needs stronger validation of
Blueprint JSON structure
Key: AMBARI-11209
URL: https://issues.apache.org/jira/browse/AMBARI-11209
Project: Ambari
Issue Type: Bug
Components: ambari-server
Affects Versions: 2.1.0
Reporter: Erik Bergenholtz
Assignee: Emil Anca
Fix For: 2.1.0
The Blueprints processor needs to have stronger validation during the process
of handling a POST for a newly-submitted Blueprint from a REST client.
The Blueprints structure requires that the "configurations" element be an array
of Maps, with each Map representing a given configuration type ("core-site",
"storm-site", etc).
Typically, the configurations structure should look like:
{code}
""configurations": [
{
"mapred-env": {
"properties":
{ "mapreduce_log_dir_prefix": "/grid/0/log/hadoop-mapreduce" }
}
},
{ "mapred-site": {
"properties":
{ "install-test-mapred-site": "install-test-mapred-site-VALUE" }
}
},
{ "yarn-env": {
"properties":
{ "yarn_log_dir_prefix": "/grid/0/log/hadoop" }
}
}
]"
The snippet above is just an example to illustrate the expected format.
Recent changes to the Blueprint's structure to handle "properties_attributes"
(such as "final") have modified the structure of a Blueprint, so that
configuration properties are now contained in a "properties" map, internal to
each configuration type. Creating new Blueprints using this newer format (the
older format is still acceptable to the processor) require that the
configuration types are still enclosed in a map, even though an internal map is
also present, which may be a source of confusion.
An example of an incorrect format would be:
""configurations": [
{
"mapred-env": {
"properties":
{ "mapreduce_log_dir_prefix": "/grid/0/log/hadoop-mapreduce" }
},
"mapred-site": {
"properties":
{ "install-test-mapred-site": "install-test-mapred-site-VALUE" }
},
"yarn-env": {
"properties":
{ "yarn_log_dir_prefix": "/grid/0/log/hadoop" }
}
]
{code}
Please note that in the example above a single map is created, with all
configuration elements being added to that entry's configuration type. This
causes a corrupted Blueprint to be stored and used for deployments, since all
configuration elements added in the original Blueprint will be registered under
one of the config types used in the original Blueprint. This will cause many
types of cluster startup failures, as the configuration elements will be added
to incorrect locations.
The underlying problem was that brackets were missing from the initial
Blueprint, although this was not immediately obvious.
The JSON parser used for the Blueprint submission will accept the incorrect
format, since it is valid JSON, but the Blueprint processor needs a higher
level of validation, such that this error is caught immediately, before any
cluster deployment is attempted.
When the Blueprint processor receives a POST submission with a new Blueprint
that includes this incorrect format for "configurations", the processor must
reject this submission, and send back a relevant error code/message, so that
the user can resolve this problem prior to attempting a cluster deployment.
This issue will not affect the older format for Blueprints, and so any
backwards-compatible Blueprints used should work fine. This issue will only
occur when new Blueprints are submitted that use the newer syntax/format for
defining configuration properties (with support for properties_attributes) as
well.
This JIRA is being created to track this issue, as it could potentially be a
user experience issue going forward, and should probably be resolved in Ambari
2.1, if possible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
