[
https://issues.apache.org/jira/browse/SAMZA-1769?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jagadish updated SAMZA-1769:
----------------------------
Description:
The first step in computing the status of a StreamApplication is to run the
ExecutionPlanner. Since the planner is run prior to everything else, any
user-level misconfiguration that can cause the planner to fail will also cause
AppRunner#status to fail.
To make the computing of AppRunner#status more robust, we should not rely on
the entire ExecutionPlan. Instead, a short-term alternative might be to
directly query the cluster-manager by providing the app-name and app-id. The
YARNJob class implements a lot of this logic - So, should offer us good
leverage here.
For the long-term, it would be preferrable to rely on the MetadataStore
abstraction in Samza to store history about runs of an application.
was:
The first step in computing the status of a StreamApplication is to run the
ExecutionPlanner. Hence, any misconfiguration by users that can cause the
planner to fail will also cause AppRunner#status to fail.
To make the computing of AppRunner#status more robust, we should not rely on
the entire ExecutionPlan. Instead, a short-term alternative might be to
directly query the cluster-manager by providing the app-name and app-id. The
YARNJob class implements a lot of this logic - So, should offer us good
leverage here.
For the long-term, it would be preferrable to rely on the MetadataStore
abstraction in Samza to store history about runs of an application.
> AppRunner#Status should not throw exceptions
> --------------------------------------------
>
> Key: SAMZA-1769
> URL: https://issues.apache.org/jira/browse/SAMZA-1769
> Project: Samza
> Issue Type: Bug
> Reporter: Jagadish
> Priority: Major
>
> The first step in computing the status of a StreamApplication is to run the
> ExecutionPlanner. Since the planner is run prior to everything else, any
> user-level misconfiguration that can cause the planner to fail will also
> cause AppRunner#status to fail.
> To make the computing of AppRunner#status more robust, we should not rely on
> the entire ExecutionPlan. Instead, a short-term alternative might be to
> directly query the cluster-manager by providing the app-name and app-id. The
> YARNJob class implements a lot of this logic - So, should offer us good
> leverage here.
> For the long-term, it would be preferrable to rely on the MetadataStore
> abstraction in Samza to store history about runs of an application.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
