[
https://issues.apache.org/jira/browse/POOL-336?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16307386#comment-16307386
]
Mark Thomas commented on POOL-336:
----------------------------------
Catching Throwable is generally a bad idea. The current code is written on the
basis that if a Throwable is thrown then that represents a fatal error so there
is no point catching it. The view was that if object factories (or the code
they call) are routinely throwing java.lang.Error then that is a bug and should
be fixed at source.
In the JDBC driver case the options are: a) raise a bug against the driver, b)
add protection in DBCP or c) add protection in POOL. My slight preference
remains for a) but I don't object to adding protection if that is the way the
community wants to go.
If we are going to go the protection route then I think it makes more sense in
POOL than in DBCP since that would be a more general fix. I don't see an
argument for providing the protection only in DBCP.
If Throwable protection is added then it needs to be added consistently - i.e.
to all locations that call external code.
Throwable protection also needs to include code along these lines:
https://svn.apache.org/viewvc/tomcat/trunk/java/org/apache/tomcat/util/ExceptionUtils.java?view=annotate#l28
to ensure that Throwables that are truly fatal are handled as such and that
Throwables that are recoverable are treated more like Exception.
> GenericObjectPool's borrowObject lock if create() fails with Error
> ------------------------------------------------------------------
>
> Key: POOL-336
> URL: https://issues.apache.org/jira/browse/POOL-336
> Project: Commons Pool
> Issue Type: Bug
> Affects Versions: 2.4.3, 2.5.0
> Environment: mariadb-connector-j
> Reporter: Wolfgang Glas
> Labels: patch
> Fix For: 2.5.1
>
> Attachments: 0001-Add-a-test-case-for-POOL-336.patch,
> 0002-Fix-POOL-336.patch
>
>
> We've spotted exactly the same problem as described in POOL-303 for a pool
> using mariadb-connector-j-2.1.2 with a pool configured to a size, which
> exactly reflects the normal quantity of used DB connections.
> After weeks of monitoring the situation, we were able to see the same problem
> as described in POOL-303. After some more drilldown efforts we concluded,
> that the problem is triggered, because mariadb-connector-j throws instances
> of Throwables derived from {{java.lang.Error}}, which are not instances of
> {{java.lang.Execption}}.
> This leads to a leak of {{makeCount}} in GenericObjectPool.create(), because
> {{makeCount}} is only decremented, when an instance of
> {{java.lang.Exception}} is thrown and not when an instance of
> {{java.langError}} is thrown.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
