[
https://issues.apache.org/jira/browse/IGNITE-18093?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-18093:
-------------------------------------
Labels: ignite-3 (was: )
> Introduce distribution zone manager for the purposes of zones configuration
> ---------------------------------------------------------------------------
>
> Key: IGNITE-18093
> URL: https://issues.apache.org/jira/browse/IGNITE-18093
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexander Lapin
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> Distribution zones manager is an ignite component that is responsible for
> both zones configuration(create/update/drop) and dataNodes calculation that
> is triggered by corresponding topology and zone modification events.
> Within the scope of current ticket it's expected that DistributionZoneManager
> will be introduced with an internal API to create, drop and alter
> distribution zones.
> h3. Definition of Done
> * New ignite component DistributionZoneManager is introduced.
> * It is inlined into ignite node start/stop flow.
> * It's possible to create/alter/drop distribution zone with the help of
> ignite internal api that technically leads to cluster configuration changes.
> h3. Implementation Notes
> 1. First of all we should introduce new root zones configuration schema:
> {code:java}
> @ConfigurationRoot(rootName = "zone", type = ConfigurationType.DISTRIBUTED)
> public class DistributionZonesConfigurationSchema {
> /** Global integer id counter. Used as an auto-increment counter to generate
> integer identifiers for distribution zone. */
> @Value(hasDefault = true)
> public int globalIdCounter = 0;
>
> /** List of configured distribution zones. */
> @NamedConfigValue
> public DistributionZoneConfigurationSchema distributionZones;
> }{code}
> and zone configuration itself:
> {code:java}
> /**
> * Distribution zone configuration schema class.
> */
> @Config
> public class DistributionZoneConfigurationSchema {
> /** Zone name. */
> @InjectedName
> public String name;
> /** Integer zone id. */
> @Immutable
> @Range(min = 1)
> @Value(hasDefault = true)
> public int zoneId = 1;
> public int dataNodesAutoAdjustScaleUp;
> public int dataNodesAutoAdjustScaleDown;
> public int dataNodesAutoAdjust;
> } {code}
> Filter and affinity function type will be added to configuration later.
> 2. It's also required to introduce DistributionZoneManager as an
> IgniteComponent that should be started and stopped during node start and stop
> flow. The proper place to insert distributionZoneManager.start() is somewhere
> after clusterCfgMgr.start().
> 3. DistributionZoneManager should expose (as in internal API) methods to
> create alter and drop distribution zones.
> 4. Aforementioned create/alter/dropZone API should be available as part of
> Ignite @TestOnly internal api.
> 5. Internally all that methods should call update configuration closures.
> 6. The only interesting part here is input data and drop validations.
> * In case of drop we should check whether there are tables binded to given
> zone. Let's say that CASCADE is not supported for now.
> * Zone name should be checked for uniqueness. Let's ask SE teem about proper
> way of doing such checks, maybe there's an annotation that will help.
> * Neither scaleUp or scaleDown are compatible with common auto adjust, so we
> should complete create result future (yep it's async) exceptionally if user
> specifies both (scaleUp || scaleDown) and common auto adjust. In case of
> alter proposed property should override non-compatible ones.
> * All-in-all some explicit validation, some configuration validators (see
> @TableValidator) and tests are expected.
> 7. Lot's of unit tests are expected to be written, that will check that
> distibutionZoneManager.create()/update()/drop() is either returns future
> completed exceptionally with proper exception or actually change the
> configuration.
> 8. Please pay attention, that within given ticket its' not expected for
> partitions and replicas to be moved inside Zones. Let it still be tables for
> now.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
