You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Alexander Lapin (Jira)" <ji...@apache.org> on 2023/05/22 12:01:00 UTC
[jira] [Commented] (IGNITE-19532) Introduce happends before relation between local meta storage safe time publication and completion of corresponding meta storage listners
[ https://issues.apache.org/jira/browse/IGNITE-19532?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17724920#comment-17724920 ]
Alexander Lapin commented on IGNITE-19532:
------------------------------------------
Without given tickets, transactions will rollback where they shouldn't (2+ nodes cluster, stable topology), meaning that we'll have flaky tests.
> Introduce happends before relation between local meta storage safe time publication and completion of corresponding meta storage listners
> -----------------------------------------------------------------------------------------------------------------------------------------
>
> Key: IGNITE-19532
> URL: https://issues.apache.org/jira/browse/IGNITE-19532
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexander Lapin
> Priority: Critical
> Labels: ignite-3
>
> h3. Motivation
> Let's assume that we have a component named LeaseTracker that locally stores leases in a map though meta storage updates
> {code:java}
> public class LeaseTracker implements PlacementDriver {
> /** Leases cache. */
> private final Map<ReplicationGroupId, Lease> leases;
> ...
> private class UpdateListener implements WatchListener {
> @Override
> public CompletableFuture<Void> onUpdate(WatchEvent event) {
> for (EntryEvent entry : event.entryEvents()) {
> ...
> Lease lease = fromBytes(msEntry.value());
> leases.put(grpId, lease);
> ...
> return completedFuture(null);
> }
> }
> ...{code}
> and we want to await lease in a meta storage safe time bounded way.
> {code:java}
> public CompletableFuture<LeaseMeta> getPrimaryReplica(ReplicationGroupId replicationGroupId, HybridTimestamp timestamp) {
> ...
> return msManager.clusterTime().waitFor(timestamp).thenApply(() -> {
> ...
> Lease lease0 = leases.get(replicationGroupId);
> if (lease.getExpirationTime().after(timestamp)) {
> return lease0;
> } else {
> return null;
> }
> ...
> }
> } {code}
> Currently that won't work properly
> because
> {code:java}
> msManager.clusterTime().waitFor(timestamp) {code}
> will be completed before local leases will be populated with data through meta storage events.
> Thus it'll be nice to complete clusterTime().waitFor(timestamp) futute (and generally speaking publish safe time) after all corresponding ms listeners are completed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)