[
https://issues.apache.org/jira/browse/IGNITE-20005?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-20005:
-------------------------------------
Description:
h3. Motivation
Please check the context in https://issues.apache.org/jira/browse/IGNITE-20033
and https://issues.apache.org/jira/browse/IGNITE-20034
After implemention writeIntentResolution coordinator path it'll be nessessary
to move commitTimestamp generation to the txn coordinator side.
h3. Definition of Done
* Easy-breezy, calculate commitTimestamp as coordinator.now().
h3. Implementation Notes
Nothing to update on the "client" side, we already have everything we need.
{code:java}
public CompletableFuture<Void> finish() {
...
HybridTimestamp commitTimestamp = commit ? clock.now() : null;
TxFinishReplicaRequest req = FACTORY.txFinishReplicaRequest()
...
.commitTimestampLong(hybridTimestampToLong(commitTimestamp)) // 2
...
} {code}
On the "server" side, we will adjust and simplify code in
org.apache.ignite.internal.table.distributed.replicator.PartitionReplicaListener#finishTransaction
{code:java}
private CompletableFuture<Object> finishTransaction(List<TablePartitionId>
aggregatedGroupIds, UUID txId, boolean commit) {
// TODO: IGNITE-17261 Timestamp from request is not using until the issue
has not been fixed (request.commitTimestamp())
var fut = new CompletableFuture<TxMeta>();
txTimestampUpdateMap.put(txId, fut);
HybridTimestamp currentTimestamp = hybridClock.now();
HybridTimestamp commitTimestamp = commit ? currentTimestamp : null; {code}
was:
h3. Motivation
It's possible that commit partition primary replica lease will expire in the
following phases:
* Before txnState change
* During txnState change. Meaning that it's actually before or after, however
we don't know an actual state before interaction.
* After txnState change.
Last part is covered with https://issues.apache.org/jira/browse/IGNITE-20002
and https://issues.apache.org/jira/browse/IGNITE-20004 So, given ticket is
about lease expiration during first and second phases. In both cases it's
required to send rollback that will change txnState to ABORTED if it wasn't
already setted to terminal state previously and return actual terminal state.
Meaning that it's possible to send rollback to a previously committed
transaction, and that rollback won't have any effect, but will return COMMITED
to the TxCoordinator and thus client.
h3. Definition of Done
* It's required to verify that we properly handle commit partition primary
replica expiration and commit partition primary replica failure by sending
rollback request to the new commit partition primary.
* It's required to slightly adjust
org.apache.ignite.internal.tx.impl.TxManagerImpl#finish along with
corresponding internals that will return actual txnState. Meaning that we
should complete user's commit successfully if we send inner rollback after
unnoticeable previous commit.
> Move commitTimestamp generation to the txn coordinator side.
> ------------------------------------------------------------
>
> Key: IGNITE-20005
> URL: https://issues.apache.org/jira/browse/IGNITE-20005
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexander Lapin
> Priority: Major
> Labels: ignite-3, transaction, transaction3_recovery
>
> h3. Motivation
> Please check the context in
> https://issues.apache.org/jira/browse/IGNITE-20033 and
> https://issues.apache.org/jira/browse/IGNITE-20034
> After implemention writeIntentResolution coordinator path it'll be nessessary
> to move commitTimestamp generation to the txn coordinator side.
> h3. Definition of Done
> * Easy-breezy, calculate commitTimestamp as coordinator.now().
> h3. Implementation Notes
> Nothing to update on the "client" side, we already have everything we need.
> {code:java}
> public CompletableFuture<Void> finish() {
> ...
> HybridTimestamp commitTimestamp = commit ? clock.now() : null;
> TxFinishReplicaRequest req = FACTORY.txFinishReplicaRequest()
> ...
> .commitTimestampLong(hybridTimestampToLong(commitTimestamp)) // 2
> ...
> } {code}
> On the "server" side, we will adjust and simplify code in
> org.apache.ignite.internal.table.distributed.replicator.PartitionReplicaListener#finishTransaction
> {code:java}
> private CompletableFuture<Object> finishTransaction(List<TablePartitionId>
> aggregatedGroupIds, UUID txId, boolean commit) {
> // TODO: IGNITE-17261 Timestamp from request is not using until the issue
> has not been fixed (request.commitTimestamp())
> var fut = new CompletableFuture<TxMeta>();
> txTimestampUpdateMap.put(txId, fut);
> HybridTimestamp currentTimestamp = hybridClock.now();
> HybridTimestamp commitTimestamp = commit ? currentTimestamp : null; {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
