[
https://issues.apache.org/jira/browse/OOZIE-2090?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robert Kanter updated OOZIE-2090:
---------------------------------
Attachment: OOZIE-2090.patch
The way wf:lastErrorNode() works is that {{DagELFunctions.setActionInfo(...)}}
checks if the action's status is ERROR and sets a variable to the node's name.
The problem is that {{ActionEndXCommand}} temporarily changes the action's
status to ERROR in this case; it gets set back to END_RETRY if there's a retry;
but the {{DagELFunctions.setActionInfo(...)}} call happens first, so it thinks
the action failed always.
The patch simply moves the call to later in {{ActionEndXCommand}} so that the
action's status can "stabilize" before setting, or not setting, it as the last
error node.
> wf:lastErrorNode does not take into account transient errors with retries
> -------------------------------------------------------------------------
>
> Key: OOZIE-2090
> URL: https://issues.apache.org/jira/browse/OOZIE-2090
> Project: Oozie
> Issue Type: Bug
> Affects Versions: 4.1.0
> Reporter: Robert Kanter
> Assignee: Robert Kanter
> Attachments: OOZIE-2090.patch
>
>
> Suppose you have a workflow where an action fails on the first try, but the
> automatic retry behavior for transient failures kicks in, and it succeeds on
> one of the later tries. Currently, the wf:lastErrorNode() EL Function will
> show that this node failed, even though it ultimately succeeded.
> We should have wf:lastErrorNode() take into account the auto-retries by not
> setting the last error node until the last retry occurs.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
