[
https://issues.apache.org/jira/browse/IGNITE-10508?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitriy Govorukhin updated IGNITE-10508:
----------------------------------------
Description:
There are cases we should trigger the checkpoint, some operations will be sure
that all operation finished before the checkpoint. It is necessary to support
the possibility of run checkpoint without waiting for the completion of the
previous checkpoint.
Solution:
Merge checkpoint pages and append write new dirty pages to a current checkpoint.
Restrictions:
Trigger new checkpoint should not wait for the previous checkpoint operation
completed.
- It should not break crash recovery mechanisms
- Only one merged is allow in the first implementation (potentially OOM, if we
will try to merge many checkpoint operations)
was:
There are cases when we should trigger the checkpoint, some operations will be
sure that all operation finished before the checkpoint. It is necessary to
support the possibility of run checkpoint without waiting for the completion of
the previous checkpoint.
Solution:
Merge checkpoint pages and append write new dirty pages to a current checkpoint.
Restrictions:
Trigger new checkpoint should not wait for the previous checkpoint operation
completed.
- It should not break crash recovery mechanisms
- Only one merged is allow in the first implementation (potentially OOM, if we
will try to merge many checkpoint operations)
> Need to support the new checkpoint feature not wait for the previous
> operation to complete
> ------------------------------------------------------------------------------------------
>
> Key: IGNITE-10508
> URL: https://issues.apache.org/jira/browse/IGNITE-10508
> Project: Ignite
> Issue Type: Improvement
> Reporter: Dmitriy Govorukhin
> Assignee: Dmitriy Govorukhin
> Priority: Major
> Fix For: 2.8
>
>
> There are cases we should trigger the checkpoint, some operations will be
> sure that all operation finished before the checkpoint. It is necessary to
> support the possibility of run checkpoint without waiting for the completion
> of the previous checkpoint.
> Solution:
> Merge checkpoint pages and append write new dirty pages to a current
> checkpoint.
> Restrictions:
> Trigger new checkpoint should not wait for the previous checkpoint operation
> completed.
> - It should not break crash recovery mechanisms
> - Only one merged is allow in the first implementation (potentially OOM, if
> we will try to merge many checkpoint operations)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
