[IGNITE-5283] Transaction recovery works incorrectly with cache store and writeThrough enabled

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.7
• Fix Version/s: 2.1
• Component/s: cache
• Assignee: Andrey N. Gura
• Reporter: Andrey N. Gura
• Created: 23/May/17 13:42
• Updated: 15/Jun/17 15:55
• Resolved: 02/Jun/17 15:03

Description

Transaction (at least optimistic) recovery on backups works incorrectly when cache store and writeThrough enabled. Changes aren't propagated to DB (it's correct) but propagated to cache if primary node left topology.

It happens because IgniteTxRemoteStateAdapter.storeUsed() method always returns false. As result incorrect code branch executes in NodeFailureTimeoutObject.onTimeout0() method (salvageTx() method should be invoked).

Steps to reproduce:

• Start two nodes A and B with cache. Cache store and writeThrough are anbled.
• From node A start optimistic transaction which executes put/update operation for key. Node A should be primary for this key.
• Kill node A when CacheStore.sessionEnd() method is invoked but changes are not propagated to underlying database (before commit on connection).
• DB doesn't contain changes but node B contains it.

Expected behaviour:

All changes should be rolled back.