[jira] [Commented] (IGNITE-21375) RO transactions should not update txnState local map
[ https://issues.apache.org/jira/browse/IGNITE-21375?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17815991#comment-17815991 ] Vladislav Pyatkov commented on IGNITE-21375: Merged 7e3605024b18cf2099110fbd038409bcbee02ee5 > RO transactions should not update txnState local map > > > Key: IGNITE-21375 > URL: https://issues.apache.org/jira/browse/IGNITE-21375 > Project: Ignite > Issue Type: Improvement >Reporter: Alexander Lapin >Assignee: Vladislav Pyatkov >Priority: Major > Labels: ignite-3, performance > Time Spent: 20m > Remaining Estimate: 0h > > h3. Motivation > On the one hand according to [~vpyatkov] updating txnstate within RO flow > brings us 20% performance drop, on the other hand RO transaction updates > corresponding state only for testing purposes, precisely in order to check > whether transaction rollback was called. See > org.apache.ignite.internal.tx.TxManager#pending for more details. > h3. Definition of Done > * Eliminate txnState local map updates within RO txn flow. > * Implement different mechanism for txn cleanup verification instead of > txnState local map based one. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (IGNITE-21375) RO transactions should not update txnState local map
[ https://issues.apache.org/jira/browse/IGNITE-21375?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17815948#comment-17815948 ] Vladislav Pyatkov commented on IGNITE-21375: This patch decreases the latency for the single-get operation by about 600 nanoseconds. This is about 7.5% of pure operation time (the pure latency is a latency without switching threads—8,4 microseconds). > RO transactions should not update txnState local map > > > Key: IGNITE-21375 > URL: https://issues.apache.org/jira/browse/IGNITE-21375 > Project: Ignite > Issue Type: Improvement >Reporter: Alexander Lapin >Assignee: Vladislav Pyatkov >Priority: Major > Labels: ignite-3, performance > Time Spent: 10m > Remaining Estimate: 0h > > h3. Motivation > On the one hand according to [~vpyatkov] updating txnstate within RO flow > brings us 20% performance drop, on the other hand RO transaction updates > corresponding state only for testing purposes, precisely in order to check > whether transaction rollback was called. See > org.apache.ignite.internal.tx.TxManager#pending for more details. > h3. Definition of Done > * Eliminate txnState local map updates within RO txn flow. > * Implement different mechanism for txn cleanup verification instead of > txnState local map based one. -- This message was sent by Atlassian Jira (v8.20.10#820010)