[SPARK-7055][SQL]Use correct ClassLoader for JDBC Driver in JDBCRDD.getConnector #5633
Conversation
|
ok to test |
|
Test build #30791 has finished for PR 5633 at commit
|
|
This is how we usually do class loading, but IIRC, there is an issue with certain JDBC drivers where they need to be loaded from the primordial classloader or else there is some security issue. /cc @tmyklebu |
|
Do the MySQL and Postgres integration tests both pass with this change? (You'll need Docker installed and you'll need to be on a Linux box to find out for yourself.) It's not a security issue per se. The trouble is that JDBC's security stuff "fails closed" and it won't be able to find the driver if it wasn't loaded with the right class loader. I believe it does this so that untrusted code can't compromise other code by polluting the driver cache. |
|
Unfortunately we can't run the docker tests on Jenkins and they cause issues with dependencies during the release so we temporarily removed them. I can try running them manually. |
|
Docker integration tests still pass. This is likely because classloaders (almost) always delegate to their parent to load, and thus even though we are allowing it to possibly find the class from the users jars it will still default to the system classloader when it can find the class. Merging to master. |
|
Thanks for fixing this! |
…etConnector Author: Vinod K C <vinod.kc@huawei.com> Closes apache#5633 from vinodkc/use_correct_classloader_driverload and squashes the following commits: 73c5380 [Vinod K C] Use correct ClassLoader for JDBC Driver
…etConnector Author: Vinod K C <vinod.kc@huawei.com> Closes apache#5633 from vinodkc/use_correct_classloader_driverload and squashes the following commits: 73c5380 [Vinod K C] Use correct ClassLoader for JDBC Driver
5 participants
No description provided.