[IGNITE-6827] Configurable rollback for long running transactions before partition exchange

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0
• Fix Version/s: 2.5
• Assignee: Alexey Scherbakov
• Reporter: Alexey Scherbakov
• Created: 03/Nov/17 12:07
• Updated: 02/Aug/19 11:13
• Resolved: 23/Apr/18 16:05

Description

Currently long running / buggy user transactions force partition exchange block on waiting for org.apache.ignite.internal.processors.cache.GridCacheSharedContext#partitionReleaseFuture, preventing all grid progress.

I suggest introducing new global flag in TransactionConfiguration, like

txRollbackTimeoutOnTopologyChange

which will rollback exchange blocking transaction after the timeout.

Still need to think what to do with other topology locking activities.

Issue Links

• blocks: IGNITE-8075 .NET: setRollbackOnTopologyChangeTimeout, withLabel, localActiveTransactions, setTxTimeoutOnPartitionMapExchange
• causes: IGNITE-11697 Suspended optimistic transaction automatically resumes to last thread after a timeout.
• incorporates: IGNITE-8000 Implicit transactions may not finish properly on unstable topology.
• incorporates: IGNITE-7585 GridDhtLockFuture related memory leak
• incorporates: IGNITE-8074 Allow changing of tx rollback timeout on exchange in runtime.
• is related to: IGNITE-7324 AssertionError Entry cannot become obsolete when DHT local candidate is added in deadlock detection suite
• is related to: IGNITE-10807 MVCC: Support long-running tx rollback on PME
• is required by: IGNITE-7910 Improve transaction debugging support
• links to: Dev.List IGNITE-6827-Review-needed
• links to: GitHub Pull Request #3094
• links to: GitHub Pull Request #3624
• links to: GitHub Pull Request #3624