Integrate hdfs services with hdfs-native

[Xuanwo]

https://github.com/Kimahriman/hdfs-native is a pure Rust implementation of an HDFS client. It might be worth considering integrating it as a new service in opendal, allowing our users to give it a try.

[shbhmrzd]

I am looking for basic hdfs operations read, write and delete in my app.
The current hdfs-sys based approach is costing in terms of image size due to the hadoop jars which need to be pulled in. If this is a pure rust implementation then it could be worth investing.
I can take this up as we already use OpenDAL in our application, and having support for hdfs-native will reduce need for an additional layer in my app.
Please let me know if this is up for grab and I can spend time on it.
Thank you!

[Xuanwo]

Hi, @shbhmrzd, I'm waiting for Kimahriman/hdfs-native#13.

[shbhmrzd]

Sure @Xuanwo
I would like to pick up OpenDAL integration when it is planned :)

[Xuanwo]

Sure @Xuanwo I would like to pick up OpenDAL integration when it is planned :)

Thanks! I will ping you here when it's ready.

[Xuanwo]

hdfs-native 0.5 has been released, let's do this!

[shbhmrzd]

Awesome
Can I take it up ?

[Xuanwo]

Awesome Can I take it up ?

Welcome! Have fun. We can add the layout first and fill the implementaion one by one.

[Xuanwo]

Thanks to @shbhmrzd and @jihuayu's efforts. We have the basic feature set now. Maybe it's time for us to establish the behavior test before adding more features.

[jihuayu]

@Xuanwo Hi, I want to add the behavior tests for it.
Is there some information about how to add the behavior test?

[Xuanwo]

Is there some information about how to add the behavior test?

Our behavior test exists at https://github.com/apache/opendal/tree/main/core/tests/behavior

We can follow the same content from hdfs for our hdfs native tests:

• fixtures: we use those fixtures to setup our services, https://github.com/apache/opendal/tree/main/fixtures/hdfs
• setup: we have a automate workflow for all services. We just need to add new setup for hdfs_native like https://github.com/apache/opendal/tree/main/.github/services/hdfs/hdfs_default

The test should be run automanticly in our CI.

[Xuanwo]

Hi, @jihuayu, do you need some help from me?

[jihuayu]

@Xuanwo Thank you!
I did encounter some trouble:
When I was preparing to write test cases, I found that the reader and writer for hdfs_native were not implemented, so I decided to implement them first.
Since I'm not very familiar with asynchronous Rust, my code has not been functioning properly. Could you help me take a look at how I should write it?
https://github.com/jihuayu/opendal/blob/f/hdfs-test/core/src/services/hdfs_native/writer.rs#L43
After running it, I found that the write never stops.

[Xuanwo]

After running it, I found that the write never stops.

FileWrtier::write is an async function, every call to f.write() will create a new future. So you will need to store this future and poll it until Ready.

For example:

opendal/core/src/raw/oio/read/range_read.rs

Lines 51 to 56 in ab52b43

	enum State<R> {
	Idle,
	SendStat(BoxedFuture<Result<RpStat>>),
	SendRead(BoxedFuture<Result<(RpRead, R)>>),
	Read(R),
	}

[jihuayu]

@Xuanwo Ohhh! I know! Thank you. I will have a try!

[Xuanwo]

Hi, @jihuayu, I did a refactor to the whole opendal's IO trait. Would you like to take another try?

[jihuayu]

@Xuanwo Thank you. I love the new trait.
I've been quite busy lately, and I'll be back in a few months.

[shbhmrzd]

@Xuanwo Can I take a swing at implementing the read and write?
Thank you!

[Xuanwo]

Can I take a swing at implementing the read and write?

Of course, have fun!