[
https://issues.apache.org/jira/browse/STORM-2819?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16258874#comment-16258874
]
Stig Rohde Døssing commented on STORM-2819:
-------------------------------------------
Kryo wasn't picked because Storm needs to serialize unknown objects. Java's
built-in serialization mechanism can do that. Kryo was picked because it's very
performant by comparison, Java serialization is very slow.
We don't have the split between native Java classes and custom classes that you
seem to think. Strings are also serialized with Kryo, the serialization for the
base classes just happen to be configured by default in Kryo so it "just works".
I think you are focusing too much on whether serialization is "native" (what do
you mean by this?). Let's say we decided to implement built-in support for
JsonObject serialization. We still have to pick a serialization format, which
would likely end up either being to/from String or via Kryo (because it
performs well). Additionally Storm would have to "know about" Gson, Jackson,
json-simple and other JSON libraries in order to provide a default
serialization implementation for that kind of JsonObject.
I just don't see the benefit here?
> Ability to natively support JSON serialization in topologies
> ------------------------------------------------------------
>
> Key: STORM-2819
> URL: https://issues.apache.org/jira/browse/STORM-2819
> Project: Apache Storm
> Issue Type: Improvement
> Components: storm-core
> Affects Versions: 1.0.2
> Environment: CentOS7, Docker
> Reporter: Anton Alfred
> Priority: Minor
> Labels: features
>
> Now that the world is moving towards NoSQL and most of the data is in JSON.
> Can a native JSON Serializer be implemented similar to support for Kryo.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
