Github user adambalogh commented on a diff in the pull request:
https://github.com/apache/spark/pull/22289#discussion_r214428426
--- Diff:
launcher/src/main/java/org/apache/spark/launcher/AbstractCommandBuilder.java ---
@@ -200,6 +200,7 @@ void addOptionString(List<String> cmd, String options) {
addToClassPath(cp, getenv("HADOOP_CONF_DIR"));
addToClassPath(cp, getenv("YARN_CONF_DIR"));
+ addToClassPath(cp, getEffectiveConfig().get("spark.yarn.conf.dir"));
--- End diff --
Yes, it is quite tricky. My expectation is that it would behave the same
way as if you pointed `HADOOP_CONF_DIR` and `YARN_CONF_DIR` to different
directories that both contain `hdfs-site.xml`. Files in `HADOOP_CONF_DIR` would
take precedence (as far as I know, nothing prevents this from happening). So
with this new config, the order of preference would be `HADOOP_CONF_DIR`,
`YARN_CONF_DIR`, then `spark.yarn.conf.dir`.
Perhaps this could be clarified in the docs, but let me know what you
think about it, I'm happy to implement other resolutions.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
