[
https://issues.apache.org/jira/browse/OOZIE-1791?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ryota Egashira updated OOZIE-1791:
----------------------------------
Description:
data pipeline customers need clean way to ignore coord job from bundle job in
runtime.
currently to make a change to failed coordinator job after bundle submission,
it is required to relaunch entire bundle after fixing coord xml or run another
bundle with new coord. OOZIE-1769 to support changing coord properties in
runtime. in many scenarios, customers want to ignore the failed coord from the
bundle first, and fix the coord, put it back to the bundle, without stopping
entire bundle job.
suggested approach is to add IGNORED status to Coordinator Job/Action. once CLI
command (oozie job -ignore <coord_job_id>) to ignore coord job issued, change
status of coord job to IGNORED (bundle action as well). Ignored coordinator
job doesn't impact the state of it's parent bundle job. e.g., after a
coordinator job failed, then ignored (suppose other coord jobs succeeded) the
bundle becomes RUNNING, not RUNNINGWITHERROR. Ignored coordinator job is
excluded from Bundle suspend/kill/rerun operation.
Only coord job in terminal state (KILLED or FAILED) can be changed to IGNORED.
PREP, RUNNING, WAITING, SUCCEEDED, SUSPENDED ones cannot be ignored. Also we
can change ignored job back to running state by using coord change command.
The same concept can be applied to coordinator action in a coordinator job.
once coord action is ignored (oozie job -ignore <coord_job_id> -action <action
number list>), it doesn't impact state of its parent coordinator job, and
excluded from coordinator job operation.
We'd like to have discussion on this approach before proceed. Any feedback
appreciated.
was:
data pipeline customers need clean way to ignore coord job from bundle job in
runtime.
currently to make a change to failed coordinator job after bundle submission,
it is required to relaunch entire bundle after fixing coord xml or run another
bundle with new coord. OOZIE-1769 to support changing coord properties in
runtime. in many scenarios, customers want to ignore the failed coord from the
bundle first, and fix the coord, put it back to the bundle, without stopping
entire bundle job.
suggested approach is to add IGNORED status to Coordinator Job/Action. once CLI
command (oozie job -ignore <coord_job_id>) to ignore coord job issued, change
status of coord job to IGNORED (bundle action as well). Ignored coordinator
job doesn't impact the state of it's parent bundle job. e.g., after a
coordinator job failed, then ignored (suppose other coord jobs succeeded) the
bundle becomes RUNNING, not RUNNINGWITHERROR. Ignored coordinator job is
excluded from Bundle suspend/kill/rerun operation.
Only coord job in terminal state (KILLED or FAILED) can be changed to IGNORED.
PREP, RUNNING, WAITING, SUCCEEDED, SUSPENDED ones cannot be ignored. Also we
can change ignored job back to running state by using coord rerun.
The same concept can be applied to coordinator action in a coordinator job.
once coord action is ignored, it doesn't impact state of its parent
coordinator job, and excluded from coordinator job operation.
We'd like to have discussion on this approach before proceed. Any feedback
appreciated.
> add IGNORED status to Coordinator Job and Action
> ------------------------------------------------
>
> Key: OOZIE-1791
> URL: https://issues.apache.org/jira/browse/OOZIE-1791
> Project: Oozie
> Issue Type: Bug
> Reporter: Ryota Egashira
> Assignee: Ryota Egashira
>
> data pipeline customers need clean way to ignore coord job from bundle job in
> runtime.
> currently to make a change to failed coordinator job after bundle submission,
> it is required to relaunch entire bundle after fixing coord xml or run
> another bundle with new coord. OOZIE-1769 to support changing coord
> properties in runtime. in many scenarios, customers want to ignore the failed
> coord from the bundle first, and fix the coord, put it back to the bundle,
> without stopping entire bundle job.
> suggested approach is to add IGNORED status to Coordinator Job/Action. once
> CLI command (oozie job -ignore <coord_job_id>) to ignore coord job issued,
> change status of coord job to IGNORED (bundle action as well). Ignored
> coordinator job doesn't impact the state of it's parent bundle job. e.g.,
> after a coordinator job failed, then ignored (suppose other coord jobs
> succeeded) the bundle becomes RUNNING, not RUNNINGWITHERROR. Ignored
> coordinator job is excluded from Bundle suspend/kill/rerun operation.
>
> Only coord job in terminal state (KILLED or FAILED) can be changed to
> IGNORED. PREP, RUNNING, WAITING, SUCCEEDED, SUSPENDED ones cannot be
> ignored. Also we can change ignored job back to running state by using coord
> change command.
> The same concept can be applied to coordinator action in a coordinator job.
> once coord action is ignored (oozie job -ignore <coord_job_id> -action
> <action number list>), it doesn't impact state of its parent coordinator
> job, and excluded from coordinator job operation.
> We'd like to have discussion on this approach before proceed. Any feedback
> appreciated.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
