[IGNITE-17845] Make Ignite Consistent Again

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Ignite Flags: Docs Required, Release Notes Required
• Assignee: Anton Vinogradov
• Reporter: Anton Vinogradov
• Created: 07/Oct/22 08:07
• Updated: 19/May/23 07:44

Description

This issue should contain issues related to:

• data and/or counters inconsistency
• obviously wrong data processing which may lead to:
  • data and/or counters inconsistency (like a current entryprocessor's algorithm at tx caches)
  • consistency guarantee level decreasing (like a mixed cache groups atomics+tx)
  • unexpected behaviour (like an atomic operations inside the transaction)
• inaccurate/incorrect consistency measurement (like a single hashcode for data and version at idle_verify)
• consistency measurement/repair tools

Issue Links

• is related to: IGNITE-17151 Inconsistent keys after deletion during rebalance
• is related to: IGNITE-15167 Control.sh should be able to fix partition inconsistency using Read Repair

Sub-Tasks

• Cluster must be able to fix the partition inconsistency on restart/node_join by itself: Resolved
• idle_verify must provide explicit data and version hashes: Resolved
• Historical rebalance must use HWM instead of LWM to seek the proper checkpoint to avoid the data loss: Resolved
• Entry Processor may cause transactional cache inconsistency: Open
• LWM may be after HWM (reserved) on primary after the node restart: Resolved
• Idle_verify must print and compare full partition counter state instead of just LWM: Resolved
• Idle_verify fails on cluster check when nodesFilter is used: Open
• Forbid mixed cache groups with both atomic and transactional caches: Open
• Change default behaviour of atomic operations inside transactions: Resolved
• TX cache update counter must deny already closed gaps re-closing: Open
• Disable partition ranges in log messages about rebalance or PME: Resolved
• TX counters with gaps must be logged on every PME: Resolved
• Topology version must be extended with topology epoch: Open
• AssertionError LWM after reserved on data insertion after the cluster restart: Resolved
• Full rebalance must be restricted when it causes any updates loss.: Open
• Remove initial counter from update counter contract.: Open
• Return highestAppliedCounter as HWM in reserved() method: In Progress
• Refactor partition counters API.: Open