[
https://issues.apache.org/jira/browse/IGNITE-17582?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-17582:
-------------------------------------
Description:
h3. Motivation
Within the scope of tx commit process, enlisted partitions are grouped together
by primary replica cluster node, however during the tx finish processing brand
new TxCleanupReplicaRequest is sent to each partition even colocated ones
within same cluster node:
{code:java}
public CompletableFuture<Void> cleanup(
ClusterNode recipientNode,
List<String> replicationGroupIds,
UUID txId,
boolean commit,
HybridTimestamp commitTimestamp
) {
// TODO: https://issues.apache.org/jira/browse/IGNITE-17582 Grouping
replica requests.
replicationGroupIds.forEach(groupId -> {
try {
replicaService.invoke(
recipientNode,
FACTORY.txCleanupReplicaRequest()
.groupId(groupId)
.txId(txId)
.commit(commit)
.commitTimestamp(commitTimestamp)
.build()
);
} catch (NodeStoppingException e) {
throw new TransactionException("Failed to perform tx cleanup, node
is stopping.");
}
});
return null;
} {code}
Seems that we may introduce multipleReplicationGroups in addition to single
ones that we already have in order to perform such collocated based requests.
h3. Definition of Done
* General solution for groupping collocated replica requests is introduced.
* TxCleanupReplicaRequests are grouped on recipient node basis.
h3. Implementation Notes
Generally speaking, it's required to:
* Introduce new interface MutipleReplicasReplicaRequest with
Collection<String> groupsId() in addition to already existing ReplicaRequest
with single String groupId();
* Implement proper multiple replicas fun out and result aggrigation(including
expcetion handling) within
org.apache.ignite.network.NetworkMessageHandler#onReceived
* Send such new TxCleanupReplicaRequests one per cluster node instead of one
per replica, in other words rework
org.apache.ignite.internal.tx.impl.TxManagerImpl#cleanup
was:
h3. Motivation
Within the scope of tx commit process, enlisted partitions are grouped together
by primary replica cluster node, however during the tx finish processing brand
new TxCleanupReplicaRequest is sent to each partition even colocated ones
within same cluster node:
{code:java}
public CompletableFuture<Void> cleanup(
ClusterNode recipientNode,
List<String> replicationGroupIds,
UUID txId,
boolean commit,
HybridTimestamp commitTimestamp
) {
// TODO: sanpwc create ticket for grouping replica requests.
replicationGroupIds.forEach(groupId -> {
try {
replicaService.invoke(
recipientNode,
FACTORY.txCleanupReplicaRequest()
.groupId(groupId)
.txId(txId)
.commit(commit)
.commitTimestamp(commitTimestamp)
.build()
);
} catch (NodeStoppingException e) {
throw new TransactionException("Failed to perform tx cleanup, node
is stopping.");
}
});
return null;
} {code}
Seems that we may introduce multipleReplicationGroups in addition to single
ones that we already have in order to perform such collocated based requests.
h3. Definition of Done
* General solution for groupping collocated replica requests is introduced.
* TxCleanupReplicaRequests are grouped on recipient node basis.
h3. Implementation Notes
Generally speaking, it's required to:
* Introduce new interface MutipleReplicasReplicaRequest with
Collection<String> groupsId() in addition to already existing ReplicaRequest
with single String groupId();
* Implement proper multiple replicas fun out and result aggrigation(including
expcetion handling) within
org.apache.ignite.network.NetworkMessageHandler#onReceived
* Send such new TxCleanupReplicaRequests one per cluster node instead of one
per replica, in other words rework
org.apache.ignite.internal.tx.impl.TxManagerImpl#cleanup
> Introduce mutipleReplica ReplicaRequests together with grouped
> TxCleanupReplicaRequest processing
> -------------------------------------------------------------------------------------------------
>
> Key: IGNITE-17582
> URL: https://issues.apache.org/jira/browse/IGNITE-17582
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexander Lapin
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> Within the scope of tx commit process, enlisted partitions are grouped
> together by primary replica cluster node, however during the tx finish
> processing brand new TxCleanupReplicaRequest is sent to each partition even
> colocated ones within same cluster node:
>
> {code:java}
> public CompletableFuture<Void> cleanup(
> ClusterNode recipientNode,
> List<String> replicationGroupIds,
> UUID txId,
> boolean commit,
> HybridTimestamp commitTimestamp
> ) {
> // TODO: https://issues.apache.org/jira/browse/IGNITE-17582 Grouping
> replica requests.
> replicationGroupIds.forEach(groupId -> {
> try {
> replicaService.invoke(
> recipientNode,
> FACTORY.txCleanupReplicaRequest()
> .groupId(groupId)
> .txId(txId)
> .commit(commit)
> .commitTimestamp(commitTimestamp)
> .build()
> );
> } catch (NodeStoppingException e) {
> throw new TransactionException("Failed to perform tx cleanup,
> node is stopping.");
> }
> });
> return null;
> } {code}
> Seems that we may introduce multipleReplicationGroups in addition to single
> ones that we already have in order to perform such collocated based requests.
> h3. Definition of Done
> * General solution for groupping collocated replica requests is introduced.
> * TxCleanupReplicaRequests are grouped on recipient node basis.
> h3. Implementation Notes
> Generally speaking, it's required to:
> * Introduce new interface MutipleReplicasReplicaRequest with
> Collection<String> groupsId() in addition to already existing ReplicaRequest
> with single String groupId();
> * Implement proper multiple replicas fun out and result
> aggrigation(including expcetion handling) within
> org.apache.ignite.network.NetworkMessageHandler#onReceived
> * Send such new TxCleanupReplicaRequests one per cluster node instead of one
> per replica, in other words rework
> org.apache.ignite.internal.tx.impl.TxManagerImpl#cleanup
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
