Roman Kondakov created IGNITE-9944:
--------------------------------------
Summary: MVCC TX: GridDhtTxAbstractEnlistFuture near nodes update
race.
Key: IGNITE-9944
URL: https://issues.apache.org/jira/browse/IGNITE-9944
Project: Ignite
Issue Type: Bug
Components: mvcc
Reporter: Roman Kondakov
Fix For: 2.7
When {{GridDhtTxAbstractEnlistFuture#skipNearNodeUpdates}} is set to true it
can cause failure on backup node during rebalancing. Possible scenario:
# First entry is enlisted before rabalance was started and therefore it's
backup node is added to {{GridDhtTxLocalAdapter#addLockTransactionNode}} in
{{GridDhtTxAbstractEnlistFuture#addToBatch}} method. Note: in this case remote
tx is not started on remote note.
# Next entry could be processed when rebalance is already started and therefore
this optimization will be skipped. In this case updated entry will be sent to
backup node as usual in DhtEnlistRequest.
# But instead of sending {{GridDhtTxQueryFirstEnlistRequest}}, the
{{GridDhtTxQueryEnlistRequest}} will be sent because near node has already been
added to tx lock nodes list.
# When this request arrives in
{{GridDhtTransactionalCacheAdapter#processDhtTxQueryEnlistRequest}}, the "Can
not find a transaction" exception will be thrown.
Possible solution: we need to send {{GridDhtTxQueryFirstEnlistRequest}} in
these cases.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
