[
https://issues.apache.org/jira/browse/IGNITE-27386?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Denis Chudov updated IGNITE-27386:
----------------------------------
Description:
*Description*
In case of exceptional tx abortion we can't restore an actual reason why it
happens. The only clue is a respond to user that may be lost. We have to store
in volatile txs' map such information within {{{}TxStateMeta{}}}:
# Exception's class type and error code.
# Trace ID.
# Exception's message.
*Motivation*
We need more information in case of exceptionally aborted transactions that
available some time for analysis.
*Definition of done*
{{TxStateMeta}} now contains the mentioned information in case of exceptional
tx abortion.
*Implementation notes*
Transaction may be aborted on coordinator or on commit partition. In case of
non-user abortion on coordinator (see InternalTableImpl#postEnlist,
TxManagerImpl#finish, TransactionStateResolver#processTxStateRequest) or on
commit partition during transaction recovery
(TxRecoveryEngine#triggerTxRecovery) tx state meta (including FINISHING state
and final states; see also exception parameters in
TransactionInflights.ReadWriteTxContext#completeFinishInProgressFuture) should
be updated with proper exception.
Also PartitionReplicaListener#appendTxCommand may throw exception "Transaction
is already finished" without proper cause, maybe similar problem in other
places.
was:
*Description*
In case of exceptional tx abortion we can't restore an actual reason why it
happens. The only clue is a respond to user that may be lost. We have to store
in volatile txs' map such information within {{{}TxStateMeta{}}}:
# Exception's class type and error code.
# Trace ID.
# Exception's message.
*Motivation*
We need more information in case of exceptionally aborted transactions that
available some time for analysis.
*Definition of done*
{{TxStateMeta}} now contains the mentioned information in case of exceptional
tx abortion.
*Implementation notes*
Transaction may be aborted on coordinator or on commit partition. In case of
non-user abortion on coordinator (see InternalTableImpl#postEnlist,
TxManagerImpl#finish, TransactionStateResolver#processTxStateRequest) or on
commit partition during transaction recovery
(TxRecoveryEngine#triggerTxRecovery) tx state meta (including FINISHING state
and final states; see also exception parameters in
TransactionInflights.ReadWriteTxContext#completeFinishInProgressFuture) should
be updated with proper exception.
> Add to TxStateMeta reasoning why a transaction was exceptionally aborted
> ------------------------------------------------------------------------
>
> Key: IGNITE-27386
> URL: https://issues.apache.org/jira/browse/IGNITE-27386
> Project: Ignite
> Issue Type: Improvement
> Reporter: Mikhail Efremov
> Assignee: Anton Laletin
> Priority: Major
> Labels: ignite-3
>
> *Description*
> In case of exceptional tx abortion we can't restore an actual reason why it
> happens. The only clue is a respond to user that may be lost. We have to
> store in volatile txs' map such information within {{{}TxStateMeta{}}}:
> # Exception's class type and error code.
> # Trace ID.
> # Exception's message.
> *Motivation*
> We need more information in case of exceptionally aborted transactions that
> available some time for analysis.
> *Definition of done*
> {{TxStateMeta}} now contains the mentioned information in case of exceptional
> tx abortion.
> *Implementation notes*
> Transaction may be aborted on coordinator or on commit partition. In case of
> non-user abortion on coordinator (see InternalTableImpl#postEnlist,
> TxManagerImpl#finish, TransactionStateResolver#processTxStateRequest) or on
> commit partition during transaction recovery
> (TxRecoveryEngine#triggerTxRecovery) tx state meta (including FINISHING state
> and final states; see also exception parameters in
> TransactionInflights.ReadWriteTxContext#completeFinishInProgressFuture)
> should be updated with proper exception.
> Also PartitionReplicaListener#appendTxCommand may throw exception
> "Transaction is already finished" without proper cause, maybe similar problem
> in other places.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
