[
https://issues.apache.org/jira/browse/DRILL-19?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13554154#comment-13554154
]
Jacques Nadeau commented on DRILL-19:
-------------------------------------
Interesting. Here are some thoughts:
- I'm not seeing the entry point into the protoschema generation. How does one
try this out?
- It seems less important to actually generate the proto text. The goal is a
canonical form of schema that we use for all our data sources. I'd really like
to leverage an existing format. I'm inclined towards a protobuf derivative
since that is what HDFS and HBase went with.
- Possible approaches are to utilize Google's DescriptorProtos/Descriptors as
the target output. Another option is to leverage what Protostuff already put
together. What would be nice about leveraging their stuff is it means we get
proto definition conversion to our schema data format (object graph) for free
using their built in compilers/transformers.
- I'm going to start working shortly on the conversion of read objects into
compact in-memory format based on protobuf. The goal is to also front this
in-memory format with RecordPointer interface that doesn't necessarily realize
the data until necessary. Once I build this, I'll work with you on getting
your scanner to output this format (as opposed to a pojo'y one).
- We should morph your iterator interface into one that looks like the one I
put together in RecordIterator in the ref interpreter. I need to add a
getSchema() option to mine. The key thing to note is how getRecordPointer is
only called once at setup and we don't do any casting.
> Build a JSON scanner that does schema discovery
> -----------------------------------------------
>
> Key: DRILL-19
> URL: https://issues.apache.org/jira/browse/DRILL-19
> Project: Apache Drill
> Issue Type: New Feature
> Reporter: Jacques Nadeau
> Assignee: Timothy Chen
>
> Build a JSON scanner that reads a file and converts it into two parts: a
> stream of records and a schema which reflects the schema of the records.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
