Description
https://github.com/confluentinc/kafka/blob/trunk/bin/kafka-run-class.sh#L128-L133
In the event CLASSPATH has not yet been populated this will result in
:$file
Normally this wouldn't be a problem however Connect's AbstractClassHearder uses ClasspathHelper.forJavaClassPath() to collect it's eligible classes.
With a leading colon you will endup with a entry for null which is expanded to the working directory.
java -cp ":" test
[]
java -cp ":Users" test
file:/Users/ryan/, file:/Users/ryan/Users
This is problematic if the script was run from the root directory since the URLs will be scanned the directories recursively. Ultimately leading to a situation where the entire FileSystem is scanned. This has been known to cause issues for some Docker installations.
Typically this is worked around by editing the the kafka-run-class script however I think we should handle this within Connect itself by excluding the root directory.