[
https://issues.apache.org/jira/browse/IGNITE-19388?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vladimir Steshin updated IGNITE-19388:
--------------------------------------
Description:
How can we request remote locks from a primary node with a transaction? Can we
send GridDhtLockRequest and reply with GridDhtLockResponse? Any tests?
We do:
- Map transaction to primary nodes where locks actually occurs.
- Remap transaction at the beginning if the topology changes.
- Stop in-progress transaction if primary leaves.
- Continue in-progress transaction if backup leaves.
- Deprecate acquiring explicit locks within transaction.
Technically, we attempt to send remote lock request further from primary
(within transaction) when primary doesn't own partition. Probably if
transaction starts at long rebalance?
However, looks like such case is not covered by tests. I've run experimental
PRs forcibly failing such remote lock processing within transaction. To be
sure, in various ways. The test have passed. But they shouldn't have.
was:
How can we request remote locks from a primary node with a transaction? Can we
send GridDhtLockRequest and reply with GridDhtLockResponse?
We do:
- Map transaction to primary nodes where locks actually occurs.
- Remap transaction at the beginning if the topology changes.
- Stop in-progress transaction if primary leaves.
- Continue in-progress transaction if backup leaves.
- Deprecate acquiring explicit locks within transaction.
Technically, we attempt to send remote lock request further from primary
(within transaction) when primary doesn't own partition. Probably if
transaction starts at long rebalance?
However, looks like such case is not covered by tests. I've run experimental
PRs forcibly failing such remote lock processing within transaction. To be
sure, in various ways. The test have passed. But they shouldn't have.
> Reachless code of out-of-primary transaction locks.
> ---------------------------------------------------
>
> Key: IGNITE-19388
> URL: https://issues.apache.org/jira/browse/IGNITE-19388
> Project: Ignite
> Issue Type: Test
> Reporter: Vladimir Steshin
> Priority: Minor
> Time Spent: 1h
> Remaining Estimate: 0h
>
> How can we request remote locks from a primary node with a transaction? Can
> we send GridDhtLockRequest and reply with GridDhtLockResponse? Any tests?
> We do:
> - Map transaction to primary nodes where locks actually occurs.
> - Remap transaction at the beginning if the topology changes.
> - Stop in-progress transaction if primary leaves.
> - Continue in-progress transaction if backup leaves.
> - Deprecate acquiring explicit locks within transaction.
>
> Technically, we attempt to send remote lock request further from primary
> (within transaction) when primary doesn't own partition. Probably if
> transaction starts at long rebalance?
>
> However, looks like such case is not covered by tests. I've run experimental
> PRs forcibly failing such remote lock processing within transaction. To be
> sure, in various ways. The test have passed. But they shouldn't have.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
