[
https://issues.apache.org/jira/browse/IGNITE-17123?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrey Mashenkov updated IGNITE-17123:
--------------------------------------
Description:
Transaction reserves partition counters on primary.
On the backup side, TxEntries must be commited with counters from the reserved
range.
However, a range of update counters, which were reserved on primary, is NOT
validated on backup. Thus means NOOP invoke operation may cause partition
counter difference on the primary and backup nodes.
1. Let's pass NOOP result of invoke operation to the backup and avoid incorrect
partition counter change on backup nodes (see DhtTxPrepareFuture).
2. Update counter can be assigned to TxEntry instantly on tx commit on Remote
node (for the WAL purposes) instead of allocating+iterating over a new
collection (see GridDistributedTxRemoteAdapter.commitIfLocked).
was:
Update counter can be assigned to TxEntry instantly on tx commit on Remote node
(for the WAL purposes) instead of allocating+iterating over a new collection
(see GridDistributedTxRemoteAdapter.commitIfLocked).
> Optimize update counter assignment on backup nodes.
> ---------------------------------------------------
>
> Key: IGNITE-17123
> URL: https://issues.apache.org/jira/browse/IGNITE-17123
> Project: Ignite
> Issue Type: Improvement
> Components: cache
> Reporter: Andrey Mashenkov
> Assignee: Andrey Mashenkov
> Priority: Minor
> Fix For: 2.14
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Transaction reserves partition counters on primary.
> On the backup side, TxEntries must be commited with counters from the
> reserved range.
> However, a range of update counters, which were reserved on primary, is NOT
> validated on backup. Thus means NOOP invoke operation may cause partition
> counter difference on the primary and backup nodes.
> 1. Let's pass NOOP result of invoke operation to the backup and avoid
> incorrect partition counter change on backup nodes (see DhtTxPrepareFuture).
> 2. Update counter can be assigned to TxEntry instantly on tx commit on Remote
> node (for the WAL purposes) instead of allocating+iterating over a new
> collection (see GridDistributedTxRemoteAdapter.commitIfLocked).
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
