You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@arrow.apache.org by Sơn Nguyễn <ho...@gmail.com> on 2025/12/17 09:47:44 UTC
Using Okhttp as an underline transport
Hello guys.
I need to use Okhttp as an underline transport for grpc, but as I checkout the Flight source code, I have a feeling that you guy only want make Flight works with Netty not Okhttp or any other transports.
I want to ask that any reason that I should not use Okhttp for Flight?
Thank you in advance
Re: Using Okhttp as an underline transport
Posted by Sơn Nguyễn <ho...@gmail.com>.
Yea, let me see
> On 29 Dec 2025, at 14:05, David Li <li...@apache.org> wrote:
>
> Ah, if the methods are being deleted, then I suppose we have no choice but to switch, if you are interested in doing that; if there's a performance issue then hopefully one of the Flight users can catch it.
>
> On Mon, Dec 29, 2025, at 13:38, Sơn Nguyễn wrote:
>> Hi David
>>
>> Thank for your answer. So I guess I can’t do anything now, It would be a lot of efforts to implement correctly and setup the benchmark.
>> One more thing, he has created a PR https://github.com/grpc/grpc-java/pull/12580 (merged) to delete some functions in ReadableBuffer that arrow is using like ReadableBuffer.readBytes(ByteBuffer dest) check the screenshot below, so I guess there would be a big change for next release.
>>
>> Regarding the Tiny url, I will stop using them next time. Thanks
>>
>> <Screenshot 2025-12-29 at 11.34.25.png>
>>
>>> On 29 Dec 2025, at 11:22, David Li <li...@apache.org> wrote:
>>>
>>> You'd probably need to ask Jacques why the code is the way it is originally, but if there's supported ways to do this now, then yes, that would be preferred. However there's not really benchmarking set up to validate that it would be equivalent, so you'd need to demonstrate that it is still as performant.
>>>
>>> (By the way, please don't use tinyurl to hide a GitHub link, it's a little suspicious...)
>>>
>>> On Mon, Dec 29, 2025, at 12:53, Sơn Nguyễn wrote:
>>>> Hi David,
>>>>
>>>> Sorry, I have to bothering you again.
>>>> But, do you have any opinions on this comment
>>>> https://tinyurl.com/3mpfpa85 from ejona86, one of the team member from
>>>> grpc-java
>>>> If what he said is true then do you welcome for a PR to update the way
>>>> we get the ByteBuffer (stop using reflection to get the ByteBuffer,...).
>>>>
>>>> Regards,
>>>> Son Nguyen
>>>>
>>>>> On 18 Dec 2025, at 08:21, Sơn Nguyễn <ho...@gmail.com> wrote:
>>>>>
>>>>> Hi David
>>>>> Thank you for your answer
>>>>>
>>>>>> On 18 Dec 2025, at 00:22, David Li <li...@apache.org> wrote:
>>>>>>
>>>>>> OkHttp is not tested. But feel free to try.
>>>>>>
>>>>>> In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
>>>>>>
>>>>>> On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
>>>>>>> Hello guys.
>>>>>>>
>>>>>>> I need to use Okhttp as an underline transport for grpc, but as I
>>>>>>> checkout the Flight source code, I have a feeling that you guy only
>>>>>>> want make Flight works with Netty not Okhttp or any other transports.
>>>>>>> I want to ask that any reason that I should not use Okhttp for Flight?
>>>>>>>
>>>>>>> Thank you in advance
>
Re: Using Okhttp as an underline transport
Posted by David Li <li...@apache.org>.
Ah, if the methods are being deleted, then I suppose we have no choice but to switch, if you are interested in doing that; if there's a performance issue then hopefully one of the Flight users can catch it.
On Mon, Dec 29, 2025, at 13:38, Sơn Nguyễn wrote:
> Hi David
>
> Thank for your answer. So I guess I can’t do anything now, It would be a lot of efforts to implement correctly and setup the benchmark.
> One more thing, he has created a PR https://github.com/grpc/grpc-java/pull/12580 (merged) to delete some functions in ReadableBuffer that arrow is using like ReadableBuffer.readBytes(ByteBuffer dest) check the screenshot below, so I guess there would be a big change for next release.
>
> Regarding the Tiny url, I will stop using them next time. Thanks
>
> Screenshot 2025-12-29 at 11.34.25.png
>
>> On 29 Dec 2025, at 11:22, David Li <li...@apache.org> wrote:
>>
>> You'd probably need to ask Jacques why the code is the way it is originally, but if there's supported ways to do this now, then yes, that would be preferred. However there's not really benchmarking set up to validate that it would be equivalent, so you'd need to demonstrate that it is still as performant.
>>
>> (By the way, please don't use tinyurl to hide a GitHub link, it's a little suspicious...)
>>
>> On Mon, Dec 29, 2025, at 12:53, Sơn Nguyễn wrote:
>>> Hi David,
>>>
>>> Sorry, I have to bothering you again.
>>> But, do you have any opinions on this comment
>>> https://tinyurl.com/3mpfpa85 from ejona86, one of the team member from
>>> grpc-java
>>> If what he said is true then do you welcome for a PR to update the way
>>> we get the ByteBuffer (stop using reflection to get the ByteBuffer,...).
>>>
>>> Regards,
>>> Son Nguyen
>>>
>>>> On 18 Dec 2025, at 08:21, Sơn Nguyễn <ho...@gmail.com> wrote:
>>>>
>>>> Hi David
>>>> Thank you for your answer
>>>>
>>>>> On 18 Dec 2025, at 00:22, David Li <li...@apache.org> wrote:
>>>>>
>>>>> OkHttp is not tested. But feel free to try.
>>>>>
>>>>> In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
>>>>>
>>>>> On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
>>>>>> Hello guys.
>>>>>>
>>>>>> I need to use Okhttp as an underline transport for grpc, but as I
>>>>>> checkout the Flight source code, I have a feeling that you guy only
>>>>>> want make Flight works with Netty not Okhttp or any other transports.
>>>>>> I want to ask that any reason that I should not use Okhttp for Flight?
>>>>>>
>>>>>> Thank you in advance
Re: Using Okhttp as an underline transport
Posted by Sơn Nguyễn <ho...@gmail.com>.
Hi David
Thank for your answer. So I guess I can’t do anything now, It would be a lot of efforts to implement correctly and setup the benchmark.
One more thing, he has created a PR https://github.com/grpc/grpc-java/pull/12580 (merged) to delete some functions in ReadableBuffer that arrow is using like ReadableBuffer.readBytes(ByteBuffer dest) check the screenshot below, so I guess there would be a big change for next release.
Regarding the Tiny url, I will stop using them next time. Thanks

> On 29 Dec 2025, at 11:22, David Li <li...@apache.org> wrote:
>
> You'd probably need to ask Jacques why the code is the way it is originally, but if there's supported ways to do this now, then yes, that would be preferred. However there's not really benchmarking set up to validate that it would be equivalent, so you'd need to demonstrate that it is still as performant.
>
> (By the way, please don't use tinyurl to hide a GitHub link, it's a little suspicious...)
>
> On Mon, Dec 29, 2025, at 12:53, Sơn Nguyễn wrote:
>> Hi David,
>>
>> Sorry, I have to bothering you again.
>> But, do you have any opinions on this comment
>> https://tinyurl.com/3mpfpa85 from ejona86, one of the team member from
>> grpc-java
>> If what he said is true then do you welcome for a PR to update the way
>> we get the ByteBuffer (stop using reflection to get the ByteBuffer,...).
>>
>> Regards,
>> Son Nguyen
>>
>>> On 18 Dec 2025, at 08:21, Sơn Nguyễn <ho...@gmail.com> wrote:
>>>
>>> Hi David
>>> Thank you for your answer
>>>
>>>> On 18 Dec 2025, at 00:22, David Li <li...@apache.org> wrote:
>>>>
>>>> OkHttp is not tested. But feel free to try.
>>>>
>>>> In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
>>>>
>>>> On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
>>>>> Hello guys.
>>>>>
>>>>> I need to use Okhttp as an underline transport for grpc, but as I
>>>>> checkout the Flight source code, I have a feeling that you guy only
>>>>> want make Flight works with Netty not Okhttp or any other transports.
>>>>> I want to ask that any reason that I should not use Okhttp for Flight?
>>>>>
>>>>> Thank you in advance
>>>
Re: Using Okhttp as an underline transport
Posted by David Li <li...@apache.org>.
You'd probably need to ask Jacques why the code is the way it is originally, but if there's supported ways to do this now, then yes, that would be preferred. However there's not really benchmarking set up to validate that it would be equivalent, so you'd need to demonstrate that it is still as performant.
(By the way, please don't use tinyurl to hide a GitHub link, it's a little suspicious...)
On Mon, Dec 29, 2025, at 12:53, Sơn Nguyễn wrote:
> Hi David,
>
> Sorry, I have to bothering you again.
> But, do you have any opinions on this comment
> https://tinyurl.com/3mpfpa85 from ejona86, one of the team member from
> grpc-java
> If what he said is true then do you welcome for a PR to update the way
> we get the ByteBuffer (stop using reflection to get the ByteBuffer,...).
>
> Regards,
> Son Nguyen
>
>> On 18 Dec 2025, at 08:21, Sơn Nguyễn <ho...@gmail.com> wrote:
>>
>> Hi David
>> Thank you for your answer
>>
>>> On 18 Dec 2025, at 00:22, David Li <li...@apache.org> wrote:
>>>
>>> OkHttp is not tested. But feel free to try.
>>>
>>> In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
>>>
>>> On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
>>>> Hello guys.
>>>>
>>>> I need to use Okhttp as an underline transport for grpc, but as I
>>>> checkout the Flight source code, I have a feeling that you guy only
>>>> want make Flight works with Netty not Okhttp or any other transports.
>>>> I want to ask that any reason that I should not use Okhttp for Flight?
>>>>
>>>> Thank you in advance
>>
Re: Using Okhttp as an underline transport
Posted by Sơn Nguyễn <ho...@gmail.com>.
Hi David,
Sorry, I have to bothering you again.
But, do you have any opinions on this comment https://tinyurl.com/3mpfpa85 from ejona86, one of the team member from grpc-java
If what he said is true then do you welcome for a PR to update the way we get the ByteBuffer (stop using reflection to get the ByteBuffer,...).
Regards,
Son Nguyen
> On 18 Dec 2025, at 08:21, Sơn Nguyễn <ho...@gmail.com> wrote:
>
> Hi David
> Thank you for your answer
>
>> On 18 Dec 2025, at 00:22, David Li <li...@apache.org> wrote:
>>
>> OkHttp is not tested. But feel free to try.
>>
>> In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
>>
>> On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
>>> Hello guys.
>>>
>>> I need to use Okhttp as an underline transport for grpc, but as I
>>> checkout the Flight source code, I have a feeling that you guy only
>>> want make Flight works with Netty not Okhttp or any other transports.
>>> I want to ask that any reason that I should not use Okhttp for Flight?
>>>
>>> Thank you in advance
>
Re: Using Okhttp as an underline transport
Posted by Sơn Nguyễn <ho...@gmail.com>.
Hi David
Thank you for your answer
> On 18 Dec 2025, at 00:22, David Li <li...@apache.org> wrote:
>
> OkHttp is not tested. But feel free to try.
>
> In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
>
> On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
>> Hello guys.
>>
>> I need to use Okhttp as an underline transport for grpc, but as I
>> checkout the Flight source code, I have a feeling that you guy only
>> want make Flight works with Netty not Okhttp or any other transports.
>> I want to ask that any reason that I should not use Okhttp for Flight?
>>
>> Thank you in advance
Re: Using Okhttp as an underline transport
Posted by David Li <li...@apache.org>.
OkHttp is not tested. But feel free to try.
In principle Flight is "just" a gRPC Protobuf definition. In practice I'm not sure that's true, unfortunately.
On Wed, Dec 17, 2025, at 18:47, Sơn Nguyễn wrote:
> Hello guys.
>
> I need to use Okhttp as an underline transport for grpc, but as I
> checkout the Flight source code, I have a feeling that you guy only
> want make Flight works with Netty not Okhttp or any other transports.
> I want to ask that any reason that I should not use Okhttp for Flight?
>
> Thank you in advance