Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

kad: Fix substream opening and dialing race #222

Merged
merged 5 commits into from
Aug 30, 2024
Merged

kad: Fix substream opening and dialing race #222

merged 5 commits into from
Aug 30, 2024

Conversation

lexnv
Copy link
Collaborator

@lexnv lexnv commented Aug 27, 2024

This PR makes the substream opening and dialing of query commands more robust.

In the past, the logic of a query action would:

  • T0; try to open a substream
  • T1; try to dial the peer

There is a race between the moment when a substream is opened and when the peer is dialed (T0 and T1), where the peer could connect to us. The race manifested after extensively running FindNode commands via subp2p-explorer bench-cli

To mitigate this race, retry to open the substream on recoverable errors from dialing (ie Error::PeerAlreadyConnected).

cc @paritytech/networking

@lexnv
kad: Make query actions robust to connection races
572cc98 
Signed-off-by: Alexandru Vasile <alexandru.vasile@parity.io>
@lexnv
find-node: Better logs on query finish
35ee1f1 
Signed-off-by: Alexandru Vasile <alexandru.vasile@parity.io>
@lexnv
kad: Refactor fault tolerant substream and dial into a dedicated fn
c4dc1ce 
Signed-off-by: Alexandru Vasile <alexandru.vasile@parity.io>
@lexnv lexnv added the bug Something isn't working label Aug 27, 2024
@lexnv lexnv self-assigned this Aug 27, 2024
dmitry-markin
dmitry-markin approved these changes Aug 28, 2024
View reviewed changes
Copy link
Collaborator

@dmitry-markin dmitry-markin left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM! Other protocols are likely also racy when opening a substream?

src/protocol/libp2p/kademlia/mod.rs Outdated Show resolved Hide resolved
@lexnv lexnv mentioned this pull request Aug 30, 2024
Open
@lexnv
Copy link
Collaborator Author

lexnv commented Aug 30, 2024

Other protocols are likely also racy when opening a substream?

After a quick look, it seems this is the only place where we tried to open a substream and then dial on failure.
I've opened this issue to try to better handle the cases where we dial a peer, however we get back a AlreadyConnected error (such as request-response and notification protocols):

@lexnv
kad: Move Error::AlreadyConnected handling into the match statement
1fbd48d 
Signed-off-by: Alexandru Vasile <alexandru.vasile@parity.io>
@lexnv lexnv merged commit 86f2a30 into master Aug 30, 2024
8 checks passed
@lexnv lexnv deleted the lexnv/kad-substream-dial branch August 30, 2024 11:22
@lexnv lexnv mentioned this pull request Sep 5, 2024
Merged
lexnv added a commit that referenced this pull request Sep 5, 2024
@lexnv
chore: Release litep2p 0.7.0 (#240)
a68b713 
## [0.7.0] - 2024-09-05

This release introduces several new features, improvements, and fixes to
the litep2p library. Key updates include enhanced error handling,
configurable connection limits, and a new API for managing public
addresses.

### [Exposing Public Addresses
API](#212)

A new `PublicAddresses` API has been added, enabling users to manage the
node's public addresses. This API allows developers to add, remove, and
retrieve public addresses, which are shared with peers through the
Identify protocol.

```rust
    // Public addresses are accessible from the main litep2p instance.
    let public_addresses = litep2p.public_addresses();

    // Add a new public address to the node.
    if let Err(err) = public_addresses.add_address(multiaddr) {
        eprintln!("Failed to add public address: {:?}", err);
    }

    // Remove a public address from the node.
    public_addresses.remove_address(&multiaddr);

    // Retrieve all public addresses of the node.
    for address in public_addresses.get_addresses() {
        println!("Public address: {}", address);
    }
```

**Breaking Change**: The Identify protocol no longer includes public
addresses in its configuration. Instead, use the new `PublicAddresses`
API.

Migration Guide:

```rust
    // Before:
    let (identify_config, identify_event_stream) = IdentifyConfig::new(
        "/substrate/1.0".to_string(),
        Some(user_agent),
        config.public_addresses,
    );

    // After:
    let (identify_config, identify_event_stream) =
        IdentifyConfig::new("/substrate/1.0".to_string(), Some(user_agent));
    // Public addresses must now be added using the `PublicAddresses` API:
    for address in config.public_addresses {
        if let Err(err) = public_addresses.add_address(address) {
            eprintln!("Failed to add public address: {:?}", err);
        }
    }
```

### [Dial Error and List Dial Failures
Event](#206)

The `DialFailure` event has been enhanced with a new `DialError` enum
for more precise error reporting when a dial attempt fails.
Additionally, a `ListDialFailures` event has been introduced, listing
all dialed addresses and their corresponding errors when multiple
addresses are involved.

Other litep2p errors, such as `ParseError`, `AddressError`, and
`NegotiationError`, have been refactored for improved error propagation.

### [Immediate Dial Error and Request-Response Rejection
Reasons](#227)

This new feature paves the way for better error handling in the
`litep2p` library and moves away from the overarching
`litep2p::error::Error` enum.

The newly added `ImmediateDialError` enum captures errors occurring
before a dial request is sent (e.g., missing peer IDs). It also enhances
the `RejectReason` enum for request-response protocols, offering more
detailed rejection reasons.


```rust
match error {
    RequestResponseError::Rejected(reason) => {
        match reason {
            RejectReason::ConnectionClosed => "connection-closed",
            RejectReason::DialFailed(Some(ImmediateDialError::AlreadyConnected)) => "already-connected",
            _ => "other",
        }
    }
    _ => "other",
}
```

### [Connection Limits](#185)

Developers can now set limits on the number of inbound and outbound
connections to manage resources and optimize performance.

```rust
    // Configure connection limits for inbound and outbound established connections.
    let litep2p_config = Config::default()
        .with_connection_limits(ConnectionLimitsConfig::default()
            .max_incoming_connections(Some(3))
            .max_outgoing_connections(Some(2))
        );
```

### [Feature Flags for Optional
Transports](#192)

The library now supports feature flags to selectively enable or disable
transport protocols. By default, only the `TCP` transport is enabled.
Optional transports include:

- `quic` - Enables QUIC transport.
- `websocket` - Enables WebSocket transport.
- `webrtc` - Enables WebRTC transport.

### [Configurable Keep-Alive
Timeout](#155)

The keep-alive timeout for connections is now configurable, providing
more control over connection lifecycles.

```rust
    // Set keep alive timeout for connections.
    let litep2p_config = Config::default()
        .with_keep_alive_timeout(Duration::from_secs(30));
```

Thanks for contributing to this @[Ma233](https://github.com/Ma233)!

### Added

- errors: Introduce immediate dial error and request-response rejection
reasons ([#227](#227))
- Expose API for `PublicAddresses`
([#212](#212))
- transport: Implement `TransportService::local_peer_id()`
([#224](#224))
- find_node: Optimize parallelism factor for slow to respond peers
([#220](#220))
- kad: Handle `ADD_PROVIDER` & `GET_PROVIDERS` network requests
([#213](#213))
- errors: Add `DialError` error and `ListDialFailures` event for better
error reporting ([#206](#206))
- kad: Add support for provider records to `MemoryStore`
([#200](#200))
- transport: Add accept_pending/reject_pending for inbound connections
and introduce inbound limits
([#194](#194))
- transport/manager: Add connection limits for inbound and outbound
established connections
([#185](#185))
- kad: Add query id to log messages
([#174](#174))

### Changed

- transport: Replace trust_dns_resolver with hickory_resolver
([#223](#223))
- crypto/noise: Generate keypair only for Curve25519
([#214](#214))
- transport: Allow manual setting of keep-alive timeout
([#155](#155))
- kad: Update connection status of an existing bucket entry
([#181](#181))
- Make transports optional
([#192](#192))

### Fixed

- kad: Fix substream opening and dialing race
([#222](#222))
- query-executor: Save the task waker on empty futures
([#219](#219))
- substream: Use write_all instead of manually writing bytes
([#217](#217))
- minor: fix tests without `websocket` feature
([#215](#215))
- Fix TCP, WebSocket, QUIC leaking connection IDs in `reject()`
([#198](#198))
- transport: Fix double lock and state overwrite on disconnected peers
([#179](#179))
- kad: Do not update memory store on incoming `GetRecordSuccess`
([#190](#190))
- transport: Reject secondary connections with different connection IDs
([#176](#176))

### Testing Done

- pulled the latest changes into Substrate
- performed warp sync in kusama

cc @paritytech/networking

---------

Signed-off-by: Alexandru Vasile <alexandru.vasile@parity.io>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@dmitry-markin dmitry-markin dmitry-markin approved these changes

Assignees

@lexnv lexnv

Labels
bug Something isn't working
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@lexnv @dmitry-markin