Remove p2pservice networkcodec trait #2388

acerone85 · 2024-10-23T17:01:08Z

Linked Issues/PRs

Description

In the P2P service, we couple together GossipSub and RequestResponse codecs in a NetworkCodec trait. In practice, this coupling is not necessary as the two codecs use different Messages and different logics for encoding/deconding messages. The only thing that they share in common is using Postcard as the data format to serialize and deserialize data, but this is not made clear in the network trait.

This PR is a rework of Codecs in the libp2p service, which deletes the NetworkCodec trait and therefore the coupling between GossipSubCodec and RequestResponseCodec. In particular:

A new DataFormat trait is defined for serializing deserializing any SerDe datatype, and the helper functions for postcard serialization and deserializationare moved to a structPostcardthat implements theDataFormat` trait,
A new BoundedCodec struct that implements the RequestResponse::Codec trait for any DataFormat supersedes the old PostcardCodec.
A new UnboundedCodec struct that implements the GossipsubCodec trait for any DataFormatsupersedes the oldPostcardCodec`,
[ ]
Parts specific to the Request/Response protocol have been moved to a new request_response::protocols module`,
The FuelBehaviour and LibP2P service have been refactored to use different codecs for Gossipsub and RequestResponse.

TODO:

Better Documentation
Better names for modules
Maybe move DataFormats to a different part of the codebase?

Checklist

Breaking changes are clearly marked as such in the PR description and changelog
New behavior is reflected in tests
The specification matches the implemented behavior (link update PR if changes are needed)

Before requesting review

I have reviewed the code myself
I have created follow-up issues caused by this PR and linked them here

After merging, notify other teams

[Add or remove entries as needed]

Rust SDK
Sway compiler
Platform documentation (for out-of-organization contributors, the person merging the PR will do this)
Someone else?

…ponses in libP2p codecs

netrome

Generally I like the decoupling you're achieving here. I think we could do some simplifications by reducing the number of traits, and for example just having one GossipSubCodec trait and oneRequestResponseReadWrite trait. Both of these could be generic over the data format if needed.

netrome · 2024-10-24T09:28:30Z

crates/services/p2p/src/codecs.rs

+trait DataFormatCodec {
+    type Error;
+    fn deserialize<'a, R: Deserialize<'a>>(
+        encoded_data: &'a [u8],
+    ) -> Result<R, Self::Error>;
+
+    fn serialize<D: Serialize>(data: &D) -> Result<Vec<u8>, Self::Error>;
+}


Are we sure there are no data formats that take parameters? I'd err on the cautious side and accept a self parameter to these methods. This would for example allow a dynamically configurable formalt such as:

enum MyFormat { Postcard, Json } While I don't know by heart when or where we'd like to use this, it feels overly restrictive to explicitly rule it out. This also eliminates the need for `PhantomData` markers in our structs, which is a bit of code smell to me.

Suggested change

trait DataFormatCodec {

type Error;

fn deserialize<'a, R: Deserialize<'a>>(

encoded_data: &'a [u8],

) -> Result<R, Self::Error>;

fn serialize<D: Serialize>(data: &D) -> Result<Vec<u8>, Self::Error>;

}

trait DataFormatCodec {

type Error;

fn deserialize<'a, R: Deserialize<'a>>(

&self,

encoded_data: &'a [u8],

) -> Result<R, Self::Error>;

fn serialize<D: Serialize>(&self, data: &D) -> Result<Vec<u8>, Self::Error>;

}

Thanks, I like the suggestion. I have made the change here: c564941 and 138716b. Note that the DataFormatCodec trait has been removed in favour of two Encode and Decode traits, following what has been done in the storage crate.

netrome · 2024-10-24T09:29:49Z

crates/services/p2p/src/codecs/bounded.rs

+
+#[derive(Debug, Clone)]
+pub struct BoundedCodec<Format> {
+    _format: PhantomData<Format>,


Following my other suggestion I'd suggest skipping the PhantomData here. If the Format is an empty struct, there's no cost to holding it directly.

Suggested change

_format: PhantomData<Format>,

_format: Format,

done, see c564941 and 138716b

netrome · 2024-10-24T09:34:21Z

crates/services/p2p/src/codecs/bounded.rs

+/// If the substream was not properly closed when dropped, the sender would instead
+/// run into a timeout waiting for the response.
+#[async_trait]
+impl<Format> request_response::Codec for BoundedCodec<Format>


The naming feels off here. A trait named Codec should have encode/decode or serialize/deserialize methods. This trait contains read/write methods and seem to be representing primarily I/O operations. I get that this comes from p2p, but perhaps we should find another name for BoundedCodec. Right now I'd expect some sort of similarity between the BoundedCodec and UnboundedCodec types, but they feel very different in what they do with the current implementations.

Perhaps I'm just not understanding this well enough right now. I'd love to hear your thoughts on this.

Agreed, I had a lot of troubles trying to find the right terminology (I also did not like BoundedCodec and UnboundedCodec. In the last version I have made the following changes:

DataFormatCodec has been replaced by Encode and Decode traits,

BoundedCodec is now a RequestResponseMessageHandler. I still don't like the Handler part in the name as it might make people think that it's something akin to a task, but I could not find anything better. The data_format field has been renamed to codec

UnboundedCodec is now GossipsubMessageHandler, and the data_format field has been renamed to codec,

The Format type parameter has been renamed to codec.

I did not rename the GossipsubCodec trait, suggestions for better names are very welcome.

Commits are ead761a and c5aa362

netrome · 2024-10-24T09:38:19Z

crates/services/p2p/src/p2p_service.rs

+        gossipsub_codec: UnboundedCodec<PostcardDataFormat>,
+        request_response_codec: BoundedCodec<PostcardDataFormat>,


Given that these are the only two instances of using these codecs, perhaps we could simplify the implementation with more specific traits. What if we just renamed UnboundedCodec -> GossipsubCodec (perhaps we only need one trait and not two traits here), and renamed BoundedCodec -> RequestResponseReadWrite (or something similar)?

Followed the suggestion: c5aa362 and ead761a

netrome · 2024-10-24T09:41:28Z

crates/services/p2p/src/codecs.rs

 use std::io;

+trait DataFormatCodec {


How about just renaming this DataFormat?

netrome · 2024-10-24T09:42:30Z

crates/services/p2p/src/codecs.rs

+trait DataFormatCodec {
+    type Error;
+    fn deserialize<'a, R: Deserialize<'a>>(
+        encoded_data: &'a [u8],


This could be generalized to a impl std::io::Read.

I'm not sure I follow completely here, do you mean changing the type of econded_data? If we do so, we need to allocate a second buffer where to read the value into. Or do you have something else in mind?

I'm not sure I follow completely here, do you mean changing the type of econded_data?

Yes, the suggestion was taking encoded_data: impl std::io::Read + 'a, instead of explicitly taking a slice here since it allows users to pass in any reader, which is more flexible. However, I see that we use AsyncRead and not Read for the network communication which would be a bit more messy so there's no immediate benefit in doing this right now anyway.

If we do so, we need to allocate a second buffer where to read the value into. Or do you have something else in mind?

That shouldn't be necessary since &[u8] also implements Read, so all current calls should still work. Or perhaps I'm misunderstanding you here.

netrome · 2024-10-24T09:43:53Z

crates/services/p2p/src/codecs.rs

+        encoded_data: &'a [u8],
+    ) -> Result<R, Self::Error>;
+
+    fn serialize<D: Serialize>(data: &D) -> Result<Vec<u8>, Self::Error>;


This could also be generalized by taking a impl std::io::Write input which it writes to. Then a convenience function serialize_to_vec could be provided with this signature.

This has been changed to reflect what's been done in the Storage crate. We now have an Encoder trait that is used as the type where the result is stored, and this is implemented for Cow<'_, [u8]>. I could in theory change this to have the input where the serialized data is written passed into the function, but this would create me trouble when I will try to merge the traits in the storage trait and this PR together (see new issue #2403)

acerone85 added 4 commits October 23, 2024 15:47

Request/Response: remove NetworkCodec trait

b6f7353

Dinstinguish between Format being used and Codec for requests and res…

529d1d6

…ponses in libP2p codecs

Move RequestResponse protocol definitions to dedicated module

a4465d5

Decouple GossibSub and RequestResponse codecs

7b858c5

acerone85 self-assigned this Oct 23, 2024

acerone85 requested review from xgreenx, Dentosal and MitchTurner as code owners October 23, 2024 17:01

acerone85 linked an issue Oct 23, 2024 that may be closed by this pull request

Remove P2PService NetworkCodec trait #2368

Open

acerone85 marked this pull request as draft October 23, 2024 17:01

acerone85 changed the title ~~2368 remove p2pservice networkcodec trait~~ Remove p2pservice networkcodec trait Oct 23, 2024

acerone85 changed the base branch from release/v0.40.0 to master October 23, 2024 17:01

netrome reviewed Oct 24, 2024

View reviewed changes

acerone85 added 11 commits October 28, 2024 14:42

Merge branch 'master' into 2368-remove-p2pservice-networkcodec-trait

f2cd720

Simplify DataFormat trait

1aa98dc

Use Encode and Decode trait in favour of DataFormat

79b507d

Rename codecs to message handlers

ead761a

Format

ed29376

Encode trait function takes &self as argument

138716b

Minor improvements

58ffe67

Decode trait function takes &self as argument

c564941

DataFormat is now Codec

c5aa362

Typo

89b022c

CHANGELOG

45bbd91

acerone85 mentioned this pull request Oct 28, 2024

Move the Encoder/Encode/Decode traits from Storage to a different crate #2403

Open

3 tasks

acerone85 added 3 commits October 28, 2024 22:32

Add issue to TODO

caf4f00

Rename lifetime

a55f695

Formatting

fc2d71d

acerone85 marked this pull request as ready for review October 29, 2024 09:48

Placate Clippy

136e6b4

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Remove p2pservice networkcodec trait #2388

Remove p2pservice networkcodec trait #2388

acerone85 commented Oct 23, 2024 •

edited

Loading

netrome left a comment

netrome Oct 24, 2024

acerone85 Oct 29, 2024 •

edited

Loading

netrome Oct 24, 2024

acerone85 Oct 29, 2024 •

edited

Loading

netrome Oct 24, 2024

acerone85 Oct 29, 2024

acerone85 Oct 29, 2024

netrome Oct 24, 2024

acerone85 Oct 29, 2024

netrome Oct 24, 2024

netrome Oct 24, 2024

acerone85 Oct 29, 2024

netrome Oct 29, 2024 •

edited

Loading

netrome Oct 24, 2024

acerone85 Oct 29, 2024

		gossipsub_codec: UnboundedCodec<PostcardDataFormat>,
		request_response_codec: BoundedCodec<PostcardDataFormat>,

Remove p2pservice networkcodec trait #2388

Are you sure you want to change the base?

Remove p2pservice networkcodec trait #2388

Conversation

acerone85 commented Oct 23, 2024 • edited Loading

Linked Issues/PRs

Description

Checklist

Before requesting review

After merging, notify other teams

netrome left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

acerone85 Oct 29, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

acerone85 Oct 29, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

netrome Oct 29, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

acerone85 commented Oct 23, 2024 •

edited

Loading

acerone85 Oct 29, 2024 •

edited

Loading

acerone85 Oct 29, 2024 •

edited

Loading

netrome Oct 29, 2024 •

edited

Loading