[jbossts-issues] [JBoss JIRA] (JBTM-3006) Tooling delete operation can left hanging remote txn branch when it was subordinate prepared participant

[Ondra Chaloupka (JIRA)]

    [ https://issues.jboss.org/browse/JBTM-3006?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13554845#comment-13554845 ] 

Ondra Chaloupka commented on JBTM-3006:
---------------------------------------

{{SubordinationManagerXAResourceOrphanFilter}} votes for leaving and I can see there the reason is that the periodic recovery gathers all the subordinate transactions under the list of imported transactions[1] while then when the transaction is checked if it's orphan[2] the transaction importer can see the transaction is in the list[3] and does not permit the orphan filtering to rollback.

If I'm not misguided I think the remove operation[4] should consider removing the the txn record from the importer _transactions[3] set. I was investigating about that but the mbean operations can't see (probably because of classpath) at the classes under 'jta' package where {{SuborindationManager}}[5] resides (where the {{TransactionImporter}}[1] could be withdrawn from).
Then I was thinking if the recovery module {{SubordinateAtomicActionRecoveryModule}} should not remove the transactions from the importer set when it adds them there[6]. But I haven't found a way[7] to do it so. I'm thinking if the orphan filter[2] could not check directly the object  store and if the transaction does not exist then it can be rolled-back. But such solution would bypass the importer which I'm not sure if it's right.

[1] https://github.com/jbosstm/narayana/blob/master/ArjunaJTA/jta/classes/com/arjuna/ats/internal/jta/transaction/arjunacore/jca/TransactionImporterImple.java#L138
[2] https://github.com/jbosstm/narayana/blob/master/ArjunaJTA/jta/classes/com/arjuna/ats/internal/jta/recovery/arjunacore/SubordinationManagerXAResourceOrphanFilter.java#L83
[3] https://github.com/jbosstm/narayana/blob/master/ArjunaJTA/jta/classes/com/arjuna/ats/internal/jta/transaction/arjunacore/jca/TransactionImporterImple.java#L161
[4] https://github.com/jbosstm/narayana/blob/master/ArjunaCore/arjuna/classes/com/arjuna/ats/arjuna/tools/osb/mbean/LogRecordWrapper.java#L236
[5] https://github.com/jbosstm/narayana/blob/master/ArjunaJTA/jta/classes/com/arjuna/ats/internal/jta/transaction/arjunacore/jca/SubordinationManager.java
[6] https://github.com/jbosstm/narayana/blob/master/ArjunaJTA/jta/classes/com/arjuna/ats/internal/jta/recovery/arjunacore/SubordinateAtomicActionRecoveryModule.java#L74
[7] https://github.com/jbosstm/narayana/blob/master/ArjunaJTA/jta/classes/com/arjuna/ats/internal/jta/recovery/arjunacore/SubordinateAtomicActionRecoveryModule.java#L93

> Tooling delete operation can left hanging remote txn branch when it was subordinate prepared participant
> --------------------------------------------------------------------------------------------------------
>
>                 Key: JBTM-3006
>                 URL: https://issues.jboss.org/browse/JBTM-3006
>             Project: JBoss Transaction Manager
>          Issue Type: Bug
>          Components: Tooling
>    Affects Versions: 5.8.0.Final
>            Reporter: Ondra Chaloupka
>            Assignee: Ondra Chaloupka
>            Priority: Minor
>         Attachments: data-subordinate.zip, jboss-cli-commands.txt, standalone-full.xml
>
>
> When administrator decides to call 'delete' in WFLY (or as jmx mbean call) and the transaction is subordinate (meaning started by another server and the context was propagated to this one) then if they are in heuristic state the transaction is removed from the narayana object store and from the RA. If the transaction is in prepared state then participant is removed from narayana object store (that could be influenced by settings of {{ignoreMBeanHeuristics}} property, see https://github.com/jbosstm/narayana/blob/master/ArjunaCore/arjuna/classes/com/arjuna/ats/arjuna/common/ObjectStoreEnvironmentBean.java#L804) but the transaction branch at the RA is left untouched - [XAResource.forget|https://docs.oracle.com/javaee/7/api/javax/transaction/xa/XAResource.html#forget-javax.transaction.xa.Xid-] is called but it fails at the participant remote branch (see [XA spec|pubs.opengroup.org/onlinepubs/009680699/toc.pdf) - the XA spec permits calling {{forget}} only on heuristically finished branches but this is still {{prepared}}.
> The final outcome is that there is removed record from Narayana object store and left untouched in-doubt (prepared) remote branch.
> Such record should be handled by recovery (ideally) the {{SubordinateJTAXAResourceOrphanFilter}} votes for rollback but then the {{SubordinationManagerXAResourceOrphanFilter}} forces to leave it as it is. Thus such record will be handing forever.
> {code}
>  TRACE [com.arjuna.ats.arjuna] (Periodic Recovery) OutputObjectState::OutputObjectState()
>  DEBUG [com.arjuna.ats.jta] (Periodic Recovery) No record found for < formatId=131077, gtrid_length=29, bqual_length=37, tx_uid=0:ffff0a280466:-51bd8cd:5aa92b2b:1c, node_name=1, branch_uid=0:ffff0a280466:758014d3:5aa92b2f:26, subordinatenodename=2, eis_name=java:/JmsXA NodeId:2bda713c-2790-11e8-8b4f-0a0027000000 >
>  DEBUG [com.arjuna.ats.jta] (Periodic Recovery) XAResourceOrphanFilter com.arjuna.ats.internal.jta.recovery.arjunacore.SubordinateJTAXAResourceOrphanFilter voted ROLLBACK
>  DEBUG [com.arjuna.ats.jta] (Periodic Recovery) subordinate node name of < formatId=131077, gtrid_length=29, bqual_length=37, tx_uid=0:ffff0a280466:-51bd8cd:5aa92b2b:1c, node_name=1, branch_uid=0:ffff0a280466:758014d3:5aa92b2f:26, subordinatenodename=2, eis_name=java:/JmsXA NodeId:2bda713c-2790-11e8-8b4f-0a0027000000 > is 2
>  TRACE [com.arjuna.ats.jta] (Periodic Recovery) TransactionImple.getStatus: javax.transaction.Status.STATUS_UNKNOWN
>  DEBUG [com.arjuna.ats.jta] (Periodic Recovery) XAResourceOrphanFilter com.arjuna.ats.internal.jta.recovery.arjunacore.SubordinationManagerXAResourceOrphanFilter voted LEAVE_ALONE
>  TRACE [com.arjuna.ats.arjuna] (Periodic Recovery) RecoveryXids isStale Check org.jboss.activemq.artemis.wildfly.integration.WildFlyActiveMQXAResourceWrapper at 3397522a 1521650139598 1521650140600 true
> {code}



--
This message was sent by Atlassian JIRA
(v7.5.0#75005)