[
https://issues.apache.org/jira/browse/IGNITE-19491?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mirza Aliev updated IGNITE-19491:
---------------------------------
Labels: dzm-reviewed ignite-3 tech-debt (was: ignite-3 tech-debt)
> Proper utilisation of a distribution zone manager's
> ZoneState#topologyAugmentationMap.
> --------------------------------------------------------------------------------------
>
> Key: IGNITE-19491
> URL: https://issues.apache.org/jira/browse/IGNITE-19491
> Project: Ignite
> Issue Type: Bug
> Reporter: Mirza Aliev
> Priority: Major
> Labels: dzm-reviewed, ignite-3, tech-debt
>
> In distribution zone manager we have several local entities that hold some
> states.
> * DistributionZoneManager.ZoneState#topologyAugmentationMap
> * DistributionZoneManager#nodesAttributes
> Currently {{DistributionZoneManager.ZoneState#topologyAugmentationMap}} is
> being cleaned only on a node, that successfully writes data nodes, which
> seems to be incorrect, because for example a node, that never writes data
> nodes, will never clean up its state.
> Need to find a proper solution for states utilisation.
> UPD:
> Proposed solution:
> ZoneState#topologyAugmentationMap should be cleared when we read scaleUp/Down
> keys in saveDataNodesToMetaStorageOnScaleUp/Down methods,
> when we retrieved this keys, it is safe to call
> {{zoneState.cleanUp(Math.min(scaleDownTriggerRevision,
> scaleUpTriggerRevision))}}, but we also need to change semantic of
> {{cleanUp}} method, so we need to call
> {{topologyAugmentationMap.headMap(toKey, false).clear()}} with false flag, to
> not include the passed key and not to remove it from the tree.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
