Add simple Ruby client example

[amoeba]

Fixes apache/arrow#40478.
Goes with open PR for Ruby server example #17.

Hi @kou, do you want to have a look? These Ruby APIs aren't familiar to me and I haven't used red-arrow before so I imagine this can be improved. The second example is nonsense right now and doesn't work but I'll look at this again tomorrow. Edit: With a fresh set of eyes, I think I fixed the streaming example, or at least made it run.

So far I've just tested this against the Ruby server PR and not others.

[kou]

(I'll add a comment about streaming approach later.)

[kou]

We need to implement arrow::ipc::StreamDecoder bindings for streaming case. arrow::ipc::StreamDecoder is used by C++ client example: https://github.com/apache/arrow-experiments/blob/main/http/get_simple/cpp/client/client.cpp

Because Net::HTTP provides read chunks to callback. It doesn't provide blocking read API. So we can't use Arrow::RecordBatchStreamReader that requires blocking read API. We need to use arrow::ipc::StreamDecoder for callback style read.

[kou]

For arrow::ipc::StreamDecoder binding: apache/arrow#40493

[amoeba]

Thanks for the review @kou. How about I remove the streaming example for now?

[ianmcook]

@amoeba if you remove the streaming example, can you copy it into a Gist and we can link to that from the README for future reference?

[amoeba]

Sure. I put it in a gist. While it works, I don't think it actually streams (I think it buffers the entire response in memory before moving on) but it's there.

[ianmcook]

Ok, thanks! I'll leave it to you to link it from the readme (or not) if you think it's worth it.

[kou]

We can remove the current streaming implementation and merge this.

BTW, the current implementation isn't streaming because res.read_body blocks until all response data received. We need to rewrite it entirely. So I think that we don't need to refer the Gist from README.

[amoeba]

Thanks @kou.

BTW, the current implementation isn't streaming because res.read_body blocks until all response data received. We need to rewrite it entirely. So I think that we don't need to refer the Gist from README.

Yep, I agree. I should've made that more clear above. I just merged your changes and am going to give them a quick test.

[amoeba]

Thanks @ianmcook. This is good to go now. I tested locally and get this when I run it at the recently-merged server example:

$ bundle exec ruby client.rb
24415 record batches received
3.51 seconds elapsed

[ianmcook]

Thanks @amoeba and @kou!

I successfully tested this against all the server examples.

While testing, I temporarily added this code at the bottom to write the table to a file:

output_file_path = "output-ruby.arrow"
output_stream = Arrow::FileOutputStream.new(output_file_path, false)
writer = Arrow::RecordBatchFileWriter.new(output_stream, schema)
writer.write_table(table)
writer.close
output_stream.close

I confirmed that all the resulting files were valid.

[kou]

While testing, I temporarily added this code at the bottom to write the table to a file:

output_file_path = "output-ruby.arrow"
output_stream = Arrow::FileOutputStream.new(output_file_path, false)
writer = Arrow::RecordBatchFileWriter.new(output_stream, schema)
writer.write_table(table)
writer.close
output_stream.close

FYI: We can write it as only table.save("output-ruby.arrows"). :-)