[jira] [Commented] (VCL-734) set requests to pending state earlier when processing

["ASF subversion and git services (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/VCL-734?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13921121#comment-13921121 ] 

ASF subversion and git services commented on VCL-734:
-----------------------------------------------------

Commit 1574592 from [~arkurth] in branch 'vcl/trunk'
[ https://svn.apache.org/r1574592 ]

VCL-734
Updated vcld to not check for cluster requests and to not update the request state to pending. The data returned from get_management_node_requests only includes reservations assigned to the running management node. Cluster reservations assigned to other MN's isn't included. Moved logic to State.pm.

Updated State.pm::initialize:
-Call reservation_failed if any of the supporting objects can't be created
-Added additional checks before a request is set to pending to detect if it was deleted or if any cluster child reservations have already failed.

Added State.pm::does_loadstate_exist_any_reservation. It is used to detect if any other cluster reservations have failed.

Moved steps which were in State.pm::DESTROY to State.pm::state_exit and added a call to state_exit in DESTROY. This ensures the steps always occur. Added a flag to state_exit so it is only run once. Added failed cluster reservation check. Added check before request state is updated to determine if it was updated by another process.

Added temporary condition to inuse.pm to skip all connection checks for cluster requests unless the previous state was reserved.

Added get_parent_reservation_id to DataStructure.pm

> set requests to pending state earlier when processing
> -----------------------------------------------------
>
>                 Key: VCL-734
>                 URL: https://issues.apache.org/jira/browse/VCL-734
>             Project: VCL
>          Issue Type: Improvement
>          Components: vcld (backend)
>            Reporter: Aaron Peeler
>            Priority: Minor
>             Fix For: 2.4
>
>
> With some of the latest code additions. It appears reservations can be picked up a second time during vcld's initial fetch of requests that need to be processed.
> The routine update_request_state gets called in the State.pm module to update a reservation to pending, which after loading several other prov/OS modules for a request, the loop in vcld can pick up the same request and try to process again.
> So for non-cluster reservations, add a call to update_request_state in vcld before forking. Cluster requests will still likely need to be updated from the State.pm module. 



--
This message was sent by Atlassian JIRA
(v6.2#6252)