[
https://issues.apache.org/jira/browse/SPARK-3474?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14985938#comment-14985938
]
Konstantin Boudnik commented on SPARK-3474:
-------------------------------------------
Actually, it seems there's still something to this problem. As of 1.5.1 I found
that starting master like this:
{code}
. /usr/lib/spark/bin/load-spark-env.sh
/usr/lib/spark/bin/spark-class org.apache.spark.deploy.master.Master
{code}
binds Master to some interface IP address instead of the hostname. Using
{{--host}} explicitly solves the issue. In this case, pointing Worker to the
spark://hostname:port is futile as they never associate with each other: Master
is bound to the IP address and not to the hostname.
> The env variable SPARK_MASTER_IP does not work
> ----------------------------------------------
>
> Key: SPARK-3474
> URL: https://issues.apache.org/jira/browse/SPARK-3474
> Project: Spark
> Issue Type: Bug
> Components: Deploy
> Affects Versions: 1.0.1
> Reporter: Chunjun Xiao
>
> There's some inconsistency regarding the env variable used to specify the
> spark master host server.
> In spark source code (MasterArguments.scala), the env variable is
> "SPARK_MASTER_HOST", while in the shell script (e.g., spark-env.sh,
> start-master.sh), it's named "SPARK_MASTER_IP".
> This will introduce an issue in some case, e.g., if spark master is started
> via "service spark-master start", which is built based on latest bigtop
> (refer to bigtop/spark-master.svc).
> In this case, "SPARK_MASTER_IP" will have no effect.
> I suggest we change SPARK_MASTER_IP in the shell script to SPARK_MASTER_HOST.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
