Package org.infinispan.statetransfer

Class TransactionSynchronizerInterceptor

java.lang.Object
org.infinispan.interceptors.BaseAsyncInterceptor
org.infinispan.statetransfer.TransactionSynchronizerInterceptor
All Implemented Interfaces:
AsyncInterceptor
public class TransactionSynchronizerInterceptor extends BaseAsyncInterceptor
With the Non-Blocking State Transfer (NBST) in place it is possible for a transactional command to be forwarded multiple times, concurrently to the same node. This interceptor makes sure that for any given transaction, the interceptor chain, post StateTransferInterceptor, would only allows a single thread to amend a transaction.

E.g. of when this situation might occur:
  • 1) Node A broadcasts PrepareCommand to nodes B, C
  • 2) Node A leaves cluster, causing new topology to be installed
  • 3) The command arrives to B and C, with lower topology than the current one
  • 4) Both B and C forward the command to node D
  • 5) D executes the two commands in parallel and finds out that A has left, therefore executing RollbackCommand>

This interceptor must placed after the logic that handles command forwarding (StateTransferInterceptor), otherwise we can end up in deadlocks when a command is forwarded in a loop to the same cache: e.g. A→B→C→A. This scenario is possible when we have chained topology changes (see ISPN-2578).

Since:
5.2
Author:
Mircea Markus