[
https://issues.apache.org/jira/browse/IGNITE-20994?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-20994:
-------------------------------------
Description:
h3. Motivation
Besides lock conflict, writeIntent resolution may also detect abandoned
transactions, thus it also should trigger initiate recovery logic. Probably
with some write intent resolution specifics.
h3. Definition of Done
Write intent resolution will initiate the recovery in case of dead tx
coordinator within commit partition path.
h3. Implementation Notes
Basically within commit partition path in case of pending state write intent
should check whether coordinator is dead (that part might be tricky because we
may lose volatile state of where coordinator is) and if it is: rollback the
transaction, and send rolled back state backwards, the one that should change
local txn state to ABORTED on initial trigger. It's not clear whether it's
required to send unlock or special sort of cleanup message to the trigger node.
> Adjust writeIntentResolution logic in order to initiate the recovery if
> coordinator is dead
> -------------------------------------------------------------------------------------------
>
> Key: IGNITE-20994
> URL: https://issues.apache.org/jira/browse/IGNITE-20994
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexander Lapin
> Priority: Major
>
> h3. Motivation
> Besides lock conflict, writeIntent resolution may also detect abandoned
> transactions, thus it also should trigger initiate recovery logic. Probably
> with some write intent resolution specifics.
> h3. Definition of Done
> Write intent resolution will initiate the recovery in case of dead tx
> coordinator within commit partition path.
> h3. Implementation Notes
> Basically within commit partition path in case of pending state write intent
> should check whether coordinator is dead (that part might be tricky because
> we may lose volatile state of where coordinator is) and if it is: rollback
> the transaction, and send rolled back state backwards, the one that should
> change local txn state to ABORTED on initial trigger. It's not clear whether
> it's required to send unlock or special sort of cleanup message to the
> trigger node.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
