[
https://issues.apache.org/jira/browse/OOZIE-2187?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15099135#comment-15099135
]
Jeffrey Olchovy commented on OOZIE-2187:
----------------------------------------
>> The global section parsing code used to add the <name-node>, <job-tracker>,
>> <job-xml>, and <configuration> from <global> to every action, even if they
>> didn't want it (e.g. email action). While not necessarily harmful, this
>> could be problematic in the future and results in larger database usage. It
>> now only adds a <name-node> and <job-tracker> if
>> requiresNameNodeJobTracker() is true and only adds a <job-xml> and
>> <configuration> if a new supportsConfigurationJobXML() is true.
This caused a regression when we upgraded our cluster from CDH 5.4.x to CDH
5.5.x. The {{fs}} action does not require NN/JT yet the documentation states
that:
"As of schema 0.4, if a name-node element is specified, then it is not
necessary for any of the paths to start with the file system URI as it is taken
from the name-node element. This is also true if the name-node is specified in
the global section (see Global Configurations)"
We were relying on global configuration to infer the scheme for paths that we
were passed to the {{fs}} action. After the upgrade, the global configuration
is no longer applicable to the {{fs}} action resulting in errors such as the
following:
{noformat}
FS001: Missing scheme in path
[/com/tapad/devicegraph/v4/graphs/2016/01/08/device-graph]
{noformat}
> Add a way to specify a default JT/RM and NN
> -------------------------------------------
>
> Key: OOZIE-2187
> URL: https://issues.apache.org/jira/browse/OOZIE-2187
> Project: Oozie
> Issue Type: New Feature
> Components: core
> Reporter: Robert Kanter
> Assignee: Robert Kanter
> Fix For: trunk
>
> Attachments: OOZIE-2187.002.patch, OOZIE-2187.003.patch,
> OOZIE-2187.patch, OOZIE-2187.patch, OOZIE-2187.patch, OOZIE-2187.patch
>
>
> Oozie is cluster agnostic, which is why we require an RM/JT and NN per action
> in your workflow (or once via the <global> section). In practice, many users
> use one Oozie server per cluster, so it's an extra burden for them to have to
> specify this all the time. It would be convenient if we added configuration
> properties to oozie-site that would let you specify a default RM/JT and NN to
> use.
> This way, these users could completely omit the {{<job-tracker>}} and
> {{<name-node>}} fields from their workflows; as an added benefit, they can
> easily update these values if they ever rename/move their RM/JT or NN. We'd
> of course still allow specifying {{<job-tracker>}} and {{<name-node>}} in
> each action and {{<global>}} to allow individual workflows or actions to
> override the default.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
