[
https://issues.apache.org/jira/browse/HADOOP-17675?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tamas Mate updated HADOOP-17675:
--------------------------------
Description:
Using LdapGroupsMapping with SSL enabled causes ClassNotFoundException when it
is called through native threads, such as Apache Impala does.
When a thread is attached to the VM, the currentThread's context classloader is
null, so when jndi internally tries to use the current thread's context
classloader to load the socket factory implementation, the
Class.forname(String, boolean, ClassLoader) method gets a null as the loader
uses the bootstrap classloader.
Meanwhile the LdapGroupsMapping class and the SslSocketFactory defined in it
is loaded by the application classloader from its classpath.
As the bootstrap classloader does not have hadoop-common in its classpath, when
a native thread tries to use/load the LdapGroupsMapping class it can't because
the bootstrap loader can't load anything from hadoop-common. The correct
solution seems to be to set the currentThread's context classloader to the
classloader of LdapGroupsMapping class before initializing the jndi internals,
and then reset to the original value after, with that we can ensure that the
behaviour of other things does not change, but this failure can be avoided as
well.
Attached the complete stacktrace to this Jira.
was:
Using LdapGroupsMapping with SSL enabled causes ClassNotFoundException when the
it is called through native threads, such as Apache Impala does.
When a thread is attached to the VM, the currentThread's context classloader is
null, so when jndi internally tries to use the current thread's context
classloader to load the socket factory implementation, the
Class.forname(String, boolean, ClassLoader) method gets a null as the loader,
and uses the bootstrap classloader.
Meanwhile the LdapGroupsMapping class and the SslSocketFactory defined in it is
loaded by the application classloader from its classpath.
As the bootstrap classloader does not have hadoop-common in its classpath, when
a native thread tries to use/load the LdapGroupsMapping class it can't because
the bootstrap loader can't load anything from hadoop-common. The correct
solution seems to be to set the currentThread's context classloader to the
classloader of LdapGroupsMapping class before initializing the jndi internals,
and then reset to the original value after, with that we can ensure that the
behaviour of other things does not change, but this failure can be avoided as
well.
Attached the complete stacktrace to this Jira.
> LdapGroupsMapping$LdapSslSocketFactory ClassNotFoundException
> -------------------------------------------------------------
>
> Key: HADOOP-17675
> URL: https://issues.apache.org/jira/browse/HADOOP-17675
> Project: Hadoop Common
> Issue Type: Improvement
> Components: common
> Affects Versions: 3.2.2
> Reporter: Tamas Mate
> Assignee: István Fajth
> Priority: Major
> Attachments: stacktrace.txt
>
>
> Using LdapGroupsMapping with SSL enabled causes ClassNotFoundException when
> it is called through native threads, such as Apache Impala does.
> When a thread is attached to the VM, the currentThread's context classloader
> is null, so when jndi internally tries to use the current thread's context
> classloader to load the socket factory implementation, the
> Class.forname(String, boolean, ClassLoader) method gets a null as the loader
> uses the bootstrap classloader.
> Meanwhile the LdapGroupsMapping class and the SslSocketFactory defined in it
> is loaded by the application classloader from its classpath.
> As the bootstrap classloader does not have hadoop-common in its classpath,
> when a native thread tries to use/load the LdapGroupsMapping class it can't
> because the bootstrap loader can't load anything from hadoop-common. The
> correct solution seems to be to set the currentThread's context classloader
> to the classloader of LdapGroupsMapping class before initializing the jndi
> internals, and then reset to the original value after, with that we can
> ensure that the behaviour of other things does not change, but this failure
> can be avoided as well.
> Attached the complete stacktrace to this Jira.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
