[jira] [Commented] (SPARK-632) Akka system names need to be normalized (since they are case-sensitive)

["Andrew Ash (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-632?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14206158#comment-14206158 ] 

Andrew Ash commented on SPARK-632:
----------------------------------

// link moved to http://doc.akka.io/docs/akka/current/additional/faq.html#what-is-the-name-of-a-remote-actor

I believe having the hostname change case will still break Spark.  But after a search of the dev and user mailing lists over the past year I haven't seen any other users with this issue.

A potential fix could be to call .toLower on the hostname in the Akka string across the cluster, but it's a little dirty to make this assumption everywhere.

Technically [hostnames ARE case insensitive|http://serverfault.com/questions/261341/is-the-hostname-case-sensitive] so Spark's behavior is wrong, but the issue is in the underlying Akka library.  This is the same underlying behavior where Akka requires that hostnames exactly match as well -- you can't use an IP address to refer to a Akka listening on a hostname -- SPARK-625.

Until Akka handles differently-cased hostnames I think can only be done with an ugly workaround.

Possibly relevant Akka issues:
- https://github.com/akka/akka/issues/15990
- https://github.com/akka/akka/issues/15007

My preference would be to close this as "Won't Fix" until it's raised again as a problem from the community.

cc [~rxin]

> Akka system names need to be normalized (since they are case-sensitive)
> -----------------------------------------------------------------------
>
>                 Key: SPARK-632
>                 URL: https://issues.apache.org/jira/browse/SPARK-632
>             Project: Spark
>          Issue Type: Bug
>            Reporter: Matt Massie
>
> The "system" name of the Akka full path is case-sensitive (see http://akka.io/faq/#what_is_the_name_of_a_remote_actor).
> Since DNS names are case-insensitive and we're using them in the "system" name, we need to normalize them (e.g. make them all lowercase).  Otherwise, users will find the "workers" will not be able to connect with the "master" even though the URI appears to be correct.
> For example, Berkeley DNS occasionally uses names e.g. foo.Berkley.EDU. If I used foo.berkeley.edu as the master adddress, the workers would write to their logs that they are connecting to foo.berkeley.edu but failed to. They never show up in the master UI.  If use the foo.Berkeley.EDU address, everything works as it should. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org