You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@atlas.apache.org by "Aaron Dossett (JIRA)" <ji...@apache.org> on 2015/10/16 18:34:05 UTC
[jira] [Commented] (ATLAS-182) Add data model for Storm topology
elements
[ https://issues.apache.org/jira/browse/ATLAS-182?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14960977#comment-14960977 ]
Aaron Dossett commented on ATLAS-182:
-------------------------------------
Comments about ATLAS-182.patch:
- The code refers to the Storm DAG. Storm topologies can contain cycles,
so it¹s not literally a DAG. I don¹t think this has any implications for
the code I see though.
- The type hierarchy looks like NODE -> SPOUT -> BOLT. I think it would
be better to have SPOUT and BOLT both be subtypes of NODE since SPOUT has
some behaviors and characteristics that a BOLT does not.
- Each NODE also has a set of configurations options associated with it.
These could be primitives (e.g. ³debug=true² or
³namenode=hdfs://nn.target.com²) or more complex configurations such as a
java class describing a file rotation policy for a bolt that writes data
to a file. I don¹t know if that level of detail is appropriate yet, but
those options are the key metadata about the NODE to reason about its
behavior.
- Some BOLTS act like ³sinks² (although Storm doesn¹t use that term) in
that the output is outside of the storm topology (e.g. writing to HDFS,
streaming to a Hive table, or enqueuing to Kafka). I don¹t know how that
should be reflected in the model. It could be a NODE with no ³outputs² but
whose configuration options would describe where the data goes, or SINK
could be a datatype in the model. After typing that out, the first option
sounds better.
- I don¹t believe KAFKA and KAFKA TOPIC need to be first class citizens in
this data model. A Kafka Spout would be represented as a SPOUT with a
JAVA_CLASS of KafkaSpout and a set of configuration options identifying
brokers, topics, etc.
- Very minor, but the patch places the .java files in addons/hive-bridge/,
I assume that should be addons/storm-bridge/?
> Add data model for Storm topology elements
> ------------------------------------------
>
> Key: ATLAS-182
> URL: https://issues.apache.org/jira/browse/ATLAS-182
> Project: Atlas
> Issue Type: Sub-task
> Affects Versions: 0.6-incubating
> Reporter: Venkatesh Seetharam
> Assignee: Venkatesh Seetharam
> Fix For: 0.6-incubating
>
> Attachments: ATLAS-182.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)