flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-5815) Add resource files configuration for Yarn Mode
Date Thu, 02 Mar 2017 09:12:45 GMT

    [ https://issues.apache.org/jira/browse/FLINK-5815?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15891880#comment-15891880

ASF GitHub Bot commented on FLINK-5815:

Github user wenlong88 commented on a diff in the pull request:

    --- Diff: flink-yarn/src/main/java/org/apache/flink/yarn/cli/FlinkYarnCLI.java ---
    @@ -85,6 +89,10 @@ public FlinkYarnCLI(String shortPrefix, String longPrefix) {
     		DETACHED = new Option(shortPrefix + "a", longPrefix + "attached", false, "Start attached");
     		ZOOKEEPER_NAMESPACE = new Option(shortPrefix + "z", longPrefix + "zookeeperNamespace",
true, "Namespace to create the Zookeeper sub-paths for high availability mode");
    +		LIB_JARS = new Option(shortPrefix + "libjars", longPrefix + "libjars", true, "Jar file
paths for job, like /home/user/lib/test.jar");
    +		FILES = new Option(shortPrefix + "files", longPrefix + "files", true, "Normal file
paths for job, like /home/user/lib/test.dict");
    +		ARCHIVES = new Option(shortPrefix + "archives", longPrefix + "archives", true, "Archived
file uris for job, like hdfs:///users/flink/common_dict#dict");
    --- End diff --
    It is totally different using the BlobStore. For clusters with BlobStore, DistributedCache
may be enough. I think in the future, we can support DistributedCache with BlobStore in the
future, and support external BlobStore structure which already run on existed framework, like
yarn distributed cache in the future

> Add resource files configuration for Yarn Mode
> ----------------------------------------------
>                 Key: FLINK-5815
>                 URL: https://issues.apache.org/jira/browse/FLINK-5815
>             Project: Flink
>          Issue Type: Improvement
>          Components: Client, YARN
>    Affects Versions: 1.3.0
>            Reporter: Wenlong Lyu
>            Assignee: Wenlong Lyu
> Currently in flink, when we want to setup a resource file to distributed cache, we need
to make the file accessible remotely by a url, which is often difficult to maintain a service
like that. What's more, when we want do add some extra jar files to job classpath, we need
to copy the jar files to blob server when submitting the jobgraph. In yarn, especially in
flip-6, the blob server is not running yet when we try to start a flink job. 
> Yarn has a efficient distributed cache implementation for application running on it,
what's more we can be easily share the files stored in hdfs in different application by distributed
cache without extra IO operations. 
> I suggest to introduce -yfiles, -ylibjars -yarchives options to FlinkYarnCLI to enable
yarn user setup their job resource files by yarn distributed cache. The options is compatible
with what is used in mapreduce, which make it easy to use for yarn user who generally has
experience on using mapreduce.

This message was sent by Atlassian JIRA

View raw message