[IGNITE-19967] Design restart of Distribution Zone Manager in accordance with the new local recovery process

• Type: Task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Labels: ignite-3
• Epic Link: Metastorage Compaction
• Ignite Flags: Docs Required, Release Notes Required
• Assignee: Mirza Aliev
• Reporter: Mirza Aliev
• Created: 12/Jul/23 13:46
• Updated: 20/Sep/23 13:44
• Resolved: 20/Sep/23 13:44

Description

In https://issues.apache.org/jira/browse/IGNITE-19061 we have provided and later implemented algorithm for restoring zones states after restart.

In https://issues.apache.org/jira/browse/IGNITE-19778 and other tickets new invariants for recovery were introduced, in particular, in the new proposal node state is recovered not until applied revision, but until some higher revision, which brings inconsistency for restoring Distribution Zone Manager because the process was based on saving states to the Vault and on mechanism with restoring from the Vault. Previously we could based our recovery on the fact, that all events, that node missed during absence, will be applied to starting from metastorage applied revision, but in the new recovery process, node receives only snapshot of the data and we cannot restore the history of data nodes related events.

In this task we need to provide a new approach, taking into account new circumstances.

Issue Links

• blocks: IGNITE-19955 Fix create zone on restart rewrites existing data nodes because of trigger key inconsistnecy
• is broken by: IGNITE-19778 Restore components state on metastorage recovery
• is depended upon by: IGNITE-18694 Recovery for DistributionZoneRebalanceEngine#metaStorageManager on DistributionZoneManager#start()
• is depended upon by: IGNITE-20326 Meta storage invokes are not completed when data nodes are recalculated in DZM
• is depended upon by: IGNITE-20050 Clean CausalityDataNodesEngine#zonesVersionedCfg which stores zones' configuration changes