[jira] [Updated] (PHOENIX-7952) UpsertSelectIT is flaky due to a connection leak check using a global counter

["Andrew Kyle Purtell (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/PHOENIX-7952?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Andrew Kyle Purtell updated PHOENIX-7952:
-----------------------------------------
    Description: {{UpsertSelectIT.testUpsertSelectOnDescToAsc}} intermittently fails in {{assertNoConnLeak}} with {{java.lang.AssertionError: expected:<0> but was:<1>}}. The test executes an auto commit server side {{DESC→ASC UPSERT INTO ... SELECT ...}} which drives {{MutationState}}'s parallel mutating iterator chain and may open auxiliary {{PhoenixConnections}} that are closed asynchronously on background threads. The test's {{@After}} hook reads a global {{GLOBAL_OPEN_PHOENIX_CONNECTIONS}} counter exactly once, immediately after the test body returns. If a background close path has not yet run its decrement by that moment, the counter still reads 1 and the hook reports a false connection leak. The race window widens when {{ENABLE_SERVER_SIDE_UPSERT_MUTATIONS}} is true, since more auxiliary connections are involved. The fix is to replace the single read with a bounded poll, for example, waiting up to 30 seconds for the counter to settle, following the same pattern used for {{MaxConcurrentConnectionsIT}}, and/or to ensure that every auxiliary connection opened on the server-side {{UPSERT SELECT}} path is decremented before control returns to the caller.

> UpsertSelectIT is flaky due to a connection leak check using a global counter
> -----------------------------------------------------------------------------
>
>                 Key: PHOENIX-7952
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-7952
>             Project: Phoenix
>          Issue Type: Sub-task
>            Reporter: Andrew Kyle Purtell
>            Assignee: Andrew Kyle Purtell
>            Priority: Major
>             Fix For: 5.4.0, 5.3.2
>
>
> {{UpsertSelectIT.testUpsertSelectOnDescToAsc}} intermittently fails in {{assertNoConnLeak}} with {{java.lang.AssertionError: expected:<0> but was:<1>}}. The test executes an auto commit server side {{DESC→ASC UPSERT INTO ... SELECT ...}} which drives {{MutationState}}'s parallel mutating iterator chain and may open auxiliary {{PhoenixConnections}} that are closed asynchronously on background threads. The test's {{@After}} hook reads a global {{GLOBAL_OPEN_PHOENIX_CONNECTIONS}} counter exactly once, immediately after the test body returns. If a background close path has not yet run its decrement by that moment, the counter still reads 1 and the hook reports a false connection leak. The race window widens when {{ENABLE_SERVER_SIDE_UPSERT_MUTATIONS}} is true, since more auxiliary connections are involved. The fix is to replace the single read with a bounded poll, for example, waiting up to 30 seconds for the counter to settle, following the same pattern used for {{MaxConcurrentConnectionsIT}}, and/or to ensure that every auxiliary connection opened on the server-side {{UPSERT SELECT}} path is decremented before control returns to the caller.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)