/build/source/nativelink-util/src/connection_manager.rs

Source
// Copyright 2024 The NativeLink Authors. All rights reserved.
//
// Licensed under the Functional Source License, Version 1.1, Apache 2.0 Future License (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//    See LICENSE file for details
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

use core::pin::Pin;
use core::task::{Context, Poll};
use core::time::Duration;
use std::collections::VecDeque;
use std::sync::Arc;

use futures::Future;
use futures::stream::{FuturesUnordered, StreamExt, unfold};
use nativelink_config::stores::Retry;
use nativelink_error::{Code, Error, make_err};
use tokio::sync::{OwnedSemaphorePermit, Semaphore, mpsc, oneshot};
use tonic::transport::{Channel, Endpoint, channel};
use tracing::{debug, error, info, warn};

use crate::background_spawn;
use crate::retry::{self, Retrier, RetryResult};

/// A helper utility that enables management of a suite of connections to an
/// upstream gRPC endpoint using Tonic.
#[derive(Debug)]
pub struct ConnectionManager {
    // The channel to request connections from the worker.
    worker_tx: mpsc::Sender<(String, oneshot::Sender<Connection>)>,
}

/// The index into `ConnectionManagerWorker::endpoints`.
type EndpointIndex = usize;
/// The identifier for a given connection to a given Endpoint, used to identify
/// when a particular connection has failed or becomes available.
type ConnectionIndex = usize;

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
struct ChannelIdentifier {
    /// The index into `ConnectionManagerWorker::endpoints` that established this
    /// Channel.
    endpoint_index: EndpointIndex,
    /// A unique identifier for this particular connection to the Endpoint.
    connection_index: ConnectionIndex,
}

/// The requests that can be made from a Connection to the
/// `ConnectionManagerWorker` such as informing it that it's been dropped or that
/// an error occurred.
enum ConnectionRequest {
    /// Notify that a Connection was dropped, if it was dropped while the
    /// connection was still pending, then return the pending Channel to be
    /// added back to the available channels.
    Dropped(Option<EstablishedChannel>),
    /// Notify that a Connection was established, return the Channel to the
    /// available channels.
    Connected(EstablishedChannel),
    /// Notify that there was a transport error on the given Channel, the bool
    /// specifies whether the connection was in the process of being established
    /// or not (i.e. whether it's been returned to available channels yet).
    Error((ChannelIdentifier, bool)),
}

/// The result of a Future that connects to a given Endpoint.  This is a tuple
/// of the index into the `ConnectionManagerWorker::endpoints` that this
/// connection is for, the iteration of the connection and the result of the
/// connection itself.
type IndexedChannel = Result<EstablishedChannel, (ChannelIdentifier, Error)>;

/// A channel that has been established to an endpoint with some metadata around
/// it to allow identification of the Channel if it errors in order to correctly
/// remove it.
#[derive(Debug, Clone)]
struct EstablishedChannel {
    /// The Channel itself that the meta data relates to.
    channel: Channel,
    /// The identifier of the channel in the worker.
    identifier: ChannelIdentifier,
}

/// The context of the worker used to manage all of the connections.  This
/// handles reconnecting to endpoints on errors and multiple connections to a
/// given endpoint.
struct ConnectionManagerWorker {
    /// The endpoints to establish Channels and the identifier of the last
    /// connection attempt to that endpoint.
    endpoints: Vec<(ConnectionIndex, Endpoint)>,
    /// The channel used to communicate between a Connection and the worker.
    connection_tx: mpsc::UnboundedSender<ConnectionRequest>,
    /// Gates the maximum number of in-flight `Connection` objects.
    /// Was an explicit `usize` counter; now an `Arc<Semaphore>` so the
    /// `OwnedSemaphorePermit` held by each `Connection` releases on
    /// drop (RAII), instead of relying on a `ConnectionRequest::Dropped`
    /// round-trip that could be lost on tonic transport errors or task
    /// aborts.
    available_connections: Arc<Semaphore>,
    /// Channels that are currently being connected.
    connecting_channels: FuturesUnordered<Pin<Box<dyn Future<Output = IndexedChannel> + Send>>>,
    /// Connected channels that are available for use.
    available_channels: VecDeque<EstablishedChannel>,
    /// Requests for a Channel when available - (reason, request)
    waiting_connections: VecDeque<(String, oneshot::Sender<Connection>)>,
    /// The retry configuration for connecting to an Endpoint, on failure will
    /// restart the retrier after a 1 second delay.
    retrier: Retrier,
}

/// The maximum number of queued requests to obtain a connection from the
/// worker before applying back pressure to the requestor.  It makes sense to
/// keep this small since it has to wait for a response anyway.
const WORKER_BACKLOG: usize = 8;

impl ConnectionManager {
    /// Create a connection manager that creates a balance list between a given
    /// set of Endpoints.  This will restrict the number of concurrent requests
    /// and automatically re-connect upon transport error.
    pub fn new(
        endpoints: impl IntoIterator<Item = Endpoint>,
        mut connections_per_endpoint: usize,
        mut max_concurrent_requests: usize,
        retry: Retry,
        jitter_fn: retry::JitterFn,
    ) -> Self {
        let (worker_tx, worker_rx) = mpsc::channel(WORKER_BACKLOG);
        // The connection messages always come from sync contexts (e.g. drop)
        // and therefore, we'd end up spawning for them if this was bounded
        // which defeats the object since there would be no backpressure
        // applied. Therefore it makes sense for this to be unbounded.
        let (connection_tx, connection_rx) = mpsc::unbounded_channel();
        let endpoints = endpoints
            .into_iter()
            .map(|endpoint| (0, endpoint))
            .collect();

        if max_concurrent_requests == 0 {
            max_concurrent_requests = Semaphore::MAX_PERMITS;
        } else {
            max_concurrent_requests = max_concurrent_requests.min(Semaphore::MAX_PERMITS);
        }
        if connections_per_endpoint == 0 {
            connections_per_endpoint = 1;
        }3
        let worker = ConnectionManagerWorker {
            endpoints,
            available_connections: Arc::new(Semaphore::new(max_concurrent_requests)),
            connection_tx,
            connecting_channels: FuturesUnordered::new(),
            available_channels: VecDeque::new(),
            waiting_connections: VecDeque::new(),
            retrier: Retrier::new(
                Arc::new(|duration| Box::pin0(tokio::time::sleep0(duration0))),
                jitter_fn,
                retry,
            ),
        };
        background_spawn!("connection_manager_worker_spawn", async move {8
            worker
                .service_requests(connections_per_endpoint, worker_rx, connection_rx)
                .await;
        });
        Self { worker_tx }
    }

    /// Get a Connection that can be used as a `tonic::Channel`, except it
    /// performs some additional counting to reconnect on error and restrict
    /// the number of concurrent connections.
    pub async fn connection(&self, reason: String) -> Result<Connection, Error> {
        let (tx, rx) = oneshot::channel();
        self.worker_tx
            .send((reason, tx))
            .await
            .map_err(|err| make_err!0(Code::Unavailable0, "Requesting a new connection: {err:?}"))?0;
        rx.await
            .map_err(|err| make_err!0(Code::Unavailable0, "Waiting for a new connection: {err:?}"))
    }
}

impl ConnectionManagerWorker {
    async fn service_requests(
        mut self,
        connections_per_endpoint: usize,
        mut worker_rx: mpsc::Receiver<(String, oneshot::Sender<Connection>)>,
        mut connection_rx: mpsc::UnboundedReceiver<ConnectionRequest>,
    ) {
        // Make the initial set of connections, connection failures will be
        // handled in the same way as future transport failures, so no need to
        // do anything special.
        for endpoint_index in 0..self.endpoints.len() {
            for _ in 0..connections_per_endpoint8 {
                self.connect_endpoint(endpoint_index, None);
            }
        }

        // The main worker loop, when select resolves one of its arms the other
        // ones are cancelled, therefore it's important that they maintain no
        // state while `await`-ing.  This is enforced through the use of
        // non-async functions to do all of the work.
        loop {
            tokio::select! {
                request220 = worker_rx.recv() => {
                    let Some((reason, request)) = request else {
                        // The ConnectionManager was dropped, shut down the
                        // worker.
                        break;
                    };
                    self.handle_worker(reason, request);
                }
                maybe_request214 = connection_rx.recv() => {
                    if let Some(request) = maybe_request {
                        self.handle_connection(request);
                    }0
                }
                maybe_connection_result12 = self.connect_next() => {
                    if let Some(connection_result) = maybe_connection_result {
                        self.handle_connected(connection_result);
                    }0
                }
            }
        }
    }

    async fn connect_next(&mut self) -> Option<IndexedChannel> {308
        if self.connecting_channels.is_empty() {
            // Make this Future never resolve, we will get cancelled by the
            // select if there's some change in state to `self` and can re-enter
            // and evaluate `connecting_channels` again.
            futures::future::pending::<()>().await;
        }
        self.connecting_channels.next().await
    }

    // This must never be made async otherwise the select may cancel it.
    fn handle_connected(&mut self, connection_result: IndexedChannel) {
        match connection_result {
            Ok(established_channel) => {
                self.available_channels.push_back(established_channel);
                self.maybe_available_connection();
            }
            // When the retrier runs out of attempts start again from the
            // beginning of the retry period.  Never want to be in a
            // situation where we give up on an Endpoint forever.
            Err((identifier, _)) => {
                self.connect_endpoint(identifier.endpoint_index, Some(identifier.connection_index));
            }
        }
    }

    fn connect_endpoint(&mut self, endpoint_index: usize, connection_index: Option<usize>) {
        let Some((current_connection_index, endpoint)) = self.endpoints.get_mut(endpoint_index)
        else {
            // Unknown endpoint, this should never happen.
            error!(?endpoint_index, "Connection to unknown endpoint requested");
            return;
        };
        let is_backoff = connection_index.is_some();
        let connection_index = connection_index.unwrap_or_else(|| {
            *current_connection_index += 1;
            *current_connection_index
        });
        if is_backoff {
            warn!(
                ?connection_index,
                endpoint = ?endpoint.uri(),
                "Connection failed, reconnecting"
            );
        } else {
            info!(
                ?connection_index,
                endpoint = ?endpoint.uri(),
                "Creating new connection"
            );
        }
        let identifier = ChannelIdentifier {
            endpoint_index,
            connection_index,
        };
        let connection_stream = unfold(endpoint.clone(), move |endpoint| async move {
            let result = endpoint.connect().await.map_err(|err| {0
                make_err!(
                    Code::Unavailable,
                    "Failed to connect to {:?}: {err:?}",
                    endpoint.uri()
                )
            });
            Some((
                result.map_or_else(RetryResult::Retry, RetryResult::Ok),
                endpoint,
            ))
        });
        let retrier = self.retrier.clone();
        self.connecting_channels.push(Box::pin(async move {
            if is_backoff {
                // Just in case the retry config is 0, then we need to
                // introduce some delay so we aren't in a hard loop.
                tokio::time::sleep(Duration::from_secs(1)).await;
            }
            retrier.retry(connection_stream).await.map_or_else(
                |err| Err((identifier, err)),
                |channel| {
                    Ok(EstablishedChannel {
                        channel,
                        identifier,
                    })
                },
            )
        }));
    }

    // This must never be made async otherwise the select may cancel it.
    fn handle_worker(&mut self, reason: String, tx: oneshot::Sender<Connection>) {
        let maybe_permit = self.available_connections.clone().try_acquire_owned().ok();
        if let Some(permit219) = maybe_permit
            && let Some(channel140) = self.available_channels.pop_front()
        {
            debug!(reason, "ConnectionManager: request running");
            self.provide_channel(channel, tx, permit);
        } else {
            debug!(
                available_permits = self.available_connections.available_permits(),
                available_channels = self.available_channels.len(),
                waiting_connections = self.waiting_connections.len(),
                reason,
                "ConnectionManager: no connection available, request queued",
            );
            self.waiting_connections.push_back((reason, tx));
        }
    }

    fn provide_channel(
        &self,
        channel: EstablishedChannel,
        tx: oneshot::Sender<Connection>,
        permit: OwnedSemaphorePermit,
    ) {
        drop(tx.send(Connection {
            tx: self.connection_tx.clone(),
            pending_channel: Some(channel.channel.clone()),
            channel,
            _permit: permit,
        }));
    }

    fn maybe_available_connection(&mut self) {
        while !self.waiting_connections.is_empty() && !self.available_channels.is_empty()89 {
            let Some(permit) = self.available_connections.clone().try_acquire_owned().ok() else {
                break;
            };
            let Some(channel) = self.available_channels.pop_front() else {
                drop(permit);
                break;
            };
            let Some((reason, tx)) = self.waiting_connections.pop_front() else {
                self.available_channels.push_front(channel);
                drop(permit);
                break;
            };
            debug!(reason, "ConnectionManager: channel available, running");
            self.provide_channel(channel, tx, permit);
        }
    }

    // This must never be made async otherwise the select may cancel it.
    fn handle_connection(&mut self, request: ConnectionRequest) {
        match request {
            ConnectionRequest::Dropped(maybe_channel) => {
                if let Some(channel) = maybe_channel {
                    self.available_channels.push_back(channel);
                }0
                self.maybe_available_connection();
            }
            ConnectionRequest::Connected(channel) => {
                self.available_channels.push_back(channel);
                self.maybe_available_connection();
            }
            // Handle a transport error on a connection by making it unavailable
            // for use and establishing a new connection to the endpoint.
            ConnectionRequest::Error((identifier, was_pending)) => {
                let should_reconnect = if was_pending {
                    true
                } else {
                    let original_length = self.available_channels.len();
                    self.available_channels
                        .retain(|channel| channel.identifier != identifier);
                    // Only reconnect if it wasn't already disconnected.
                    original_length != self.available_channels.len()
                };
                if should_reconnect {
                    self.connect_endpoint(identifier.endpoint_index, None);
                }
            }
        }
    }
}

/// An instance of this is obtained for every communication with the gGRPC
/// service.  This handles the permit for limiting concurrency, and also
/// re-connecting the underlying channel on error.  It depends on users
/// reporting all errors.
/// NOTE: This should never be cloneable because its lifetime is linked to the
///       semaphore permit it carries — `_permit` is released exactly once,
///       when the `Connection` drops.
#[derive(Debug)]
pub struct Connection {
    /// Communication with `ConnectionManagerWorker` to inform about transport
    /// errors and when the Connection is dropped.
    tx: mpsc::UnboundedSender<ConnectionRequest>,
    /// If set, the Channel that will be returned to the worker when connection
    /// completes (success or failure) or when the Connection is dropped if that
    /// happens before connection completes.
    pending_channel: Option<Channel>,
    /// The identifier to send to `tx`.
    channel: EstablishedChannel,
    _permit: OwnedSemaphorePermit,
}

impl Drop for Connection {
    fn drop(&mut self) {
        let pending_channel = self
            .pending_channel
            .take()
            .map(|channel| EstablishedChannel {
                channel,
                identifier: self.channel.identifier,
            });
        drop(self.tx.send(ConnectionRequest::Dropped(pending_channel)));
    }
}

/// A wrapper around the `channel::ResponseFuture` that forwards errors to the `tx`.
pub struct ResponseFuture {
    /// The wrapped future that actually does the work.
    inner: channel::ResponseFuture,
    /// Communication with `ConnectionManagerWorker` to inform about transport
    /// errors.
    connection_tx: mpsc::UnboundedSender<ConnectionRequest>,
    /// The identifier to send to `connection_tx` on a transport error.
    identifier: ChannelIdentifier,
}

impl core::fmt::Debug for ResponseFuture {
    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
        f.debug_struct("ResponseFuture")
            .field("inner", &self.inner)
            .field("connection_tx", &self.connection_tx)
            .field("identifier", &self.identifier)
            .finish()
    }
}

/// This is mostly copied from `tonic::transport::channel` except it wraps it
/// to allow messaging about connection success and failure.
impl tonic::codegen::Service<tonic::codegen::http::Request<tonic::body::Body>> for Connection {
    type Response = tonic::codegen::http::Response<tonic::body::Body>;
    type Error = tonic::transport::Error;
    type Future = ResponseFuture;

    fn poll_ready(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
        let result = self.channel.channel.poll_ready(cx);
        if let Poll::Ready(result) = &result {
            match result {
                Ok(()) => {
                    if let Some(pending_channel) = self.pending_channel.take() {
                        drop(
                            self.tx
                                .send(ConnectionRequest::Connected(EstablishedChannel {
                                    channel: pending_channel,
                                    identifier: self.channel.identifier,
                                })),
                        );
                    }0
                }
                Err(err) => {
                    debug!(?err, "Error while creating connection on channel");
                    drop(self.tx.send(ConnectionRequest::Error((
                        self.channel.identifier,
                        self.pending_channel.take().is_some(),
                    ))));
                }
            }
        }
        result
    }

    fn call(&mut self, request: tonic::codegen::http::Request<tonic::body::Body>) -> Self::Future {
        ResponseFuture {
            inner: self.channel.channel.call(request),
            connection_tx: self.tx.clone(),
            identifier: self.channel.identifier,
        }
    }
}

/// This is mostly copied from `tonic::transport::channel` except it wraps it
/// to allow messaging about connection failure.
impl Future for ResponseFuture {
    type Output =
        Result<tonic::codegen::http::Response<tonic::body::Body>, tonic::transport::Error>;

    fn poll(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
        let result = Pin::new(&mut self.inner).poll(cx);
        if let Poll::Ready(Err(_)) = &result {
            drop(
                self.connection_tx
                    .send(ConnectionRequest::Error((self.identifier, false))),
            );
        }
        result
    }
}

Coverage Report

Created: 2026-06-04 10:48

Line	Count	Source
1		// Copyright 2024 The NativeLink Authors. All rights reserved.
2		//
3		// Licensed under the Functional Source License, Version 1.1, Apache 2.0 Future License (the "License");
4		// you may not use this file except in compliance with the License.
5		// You may obtain a copy of the License at
6		//
7		// See LICENSE file for details
8		//
9		// Unless required by applicable law or agreed to in writing, software
10		// distributed under the License is distributed on an "AS IS" BASIS,
11		// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12		// See the License for the specific language governing permissions and
13		// limitations under the License.
14
15		use core::pin::Pin;
16		use core::task::{Context, Poll};
17		use core::time::Duration;
18		use std::collections::VecDeque;
19		use std::sync::Arc;
20
21		use futures::Future;
22		use futures::stream::{FuturesUnordered, StreamExt, unfold};
23		use nativelink_config::stores::Retry;
24		use nativelink_error::{Code, Error, make_err};
25		use tokio::sync::{OwnedSemaphorePermit, Semaphore, mpsc, oneshot};
26		use tonic::transport::{Channel, Endpoint, channel};
27		use tracing::{debug, error, info, warn};
28
29		use crate::background_spawn;
30		use crate::retry::{self, Retrier, RetryResult};
31
32		/// A helper utility that enables management of a suite of connections to an
33		/// upstream gRPC endpoint using Tonic.
34		#[derive(Debug)]
35		pub struct ConnectionManager {
36		// The channel to request connections from the worker.
37		worker_tx: mpsc::Sender<(String, oneshot::Sender<Connection>)>,
38		}
39
40		/// The index into `ConnectionManagerWorker::endpoints`.
41		type EndpointIndex = usize;
42		/// The identifier for a given connection to a given Endpoint, used to identify
43		/// when a particular connection has failed or becomes available.
44		type ConnectionIndex = usize;
45
46		#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
47		struct ChannelIdentifier {
48		/// The index into `ConnectionManagerWorker::endpoints` that established this
49		/// Channel.
50		endpoint_index: EndpointIndex,
51		/// A unique identifier for this particular connection to the Endpoint.
52		connection_index: ConnectionIndex,
53		}
54
55		/// The requests that can be made from a Connection to the
56		/// `ConnectionManagerWorker` such as informing it that it's been dropped or that
57		/// an error occurred.
58		enum ConnectionRequest {
59		/// Notify that a Connection was dropped, if it was dropped while the
60		/// connection was still pending, then return the pending Channel to be
61		/// added back to the available channels.
62		Dropped(Option<EstablishedChannel>),
63		/// Notify that a Connection was established, return the Channel to the
64		/// available channels.
65		Connected(EstablishedChannel),
66		/// Notify that there was a transport error on the given Channel, the bool
67		/// specifies whether the connection was in the process of being established
68		/// or not (i.e. whether it's been returned to available channels yet).
69		Error((ChannelIdentifier, bool)),
70		}
71
72		/// The result of a Future that connects to a given Endpoint. This is a tuple
73		/// of the index into the `ConnectionManagerWorker::endpoints` that this
74		/// connection is for, the iteration of the connection and the result of the
75		/// connection itself.
76		type IndexedChannel = Result<EstablishedChannel, (ChannelIdentifier, Error)>;
77
78		/// A channel that has been established to an endpoint with some metadata around
79		/// it to allow identification of the Channel if it errors in order to correctly
80		/// remove it.
81		#[derive(Debug, Clone)]
82		struct EstablishedChannel {
83		/// The Channel itself that the meta data relates to.
84		channel: Channel,
85		/// The identifier of the channel in the worker.
86		identifier: ChannelIdentifier,
87		}
88
89		/// The context of the worker used to manage all of the connections. This
90		/// handles reconnecting to endpoints on errors and multiple connections to a
91		/// given endpoint.
92		struct ConnectionManagerWorker {
93		/// The endpoints to establish Channels and the identifier of the last
94		/// connection attempt to that endpoint.
95		endpoints: Vec<(ConnectionIndex, Endpoint)>,
96		/// The channel used to communicate between a Connection and the worker.
97		connection_tx: mpsc::UnboundedSender<ConnectionRequest>,
98		/// Gates the maximum number of in-flight `Connection` objects.
99		/// Was an explicit `usize` counter; now an `Arc<Semaphore>` so the
100		/// `OwnedSemaphorePermit` held by each `Connection` releases on
101		/// drop (RAII), instead of relying on a `ConnectionRequest::Dropped`
102		/// round-trip that could be lost on tonic transport errors or task
103		/// aborts.
104		available_connections: Arc<Semaphore>,
105		/// Channels that are currently being connected.
106		connecting_channels: FuturesUnordered<Pin<Box<dyn Future<Output = IndexedChannel> + Send>>>,
107		/// Connected channels that are available for use.
108		available_channels: VecDeque<EstablishedChannel>,
109		/// Requests for a Channel when available - (reason, request)
110		waiting_connections: VecDeque<(String, oneshot::Sender<Connection>)>,
111		/// The retry configuration for connecting to an Endpoint, on failure will
112		/// restart the retrier after a 1 second delay.
113		retrier: Retrier,
114		}
115
116		/// The maximum number of queued requests to obtain a connection from the
117		/// worker before applying back pressure to the requestor. It makes sense to
118		/// keep this small since it has to wait for a response anyway.
119		const WORKER_BACKLOG: usize = 8;
120
121		impl ConnectionManager {
122		/// Create a connection manager that creates a balance list between a given
123		/// set of Endpoints. This will restrict the number of concurrent requests
124		/// and automatically re-connect upon transport error.
125	9	pub fn new(
126	9	endpoints: impl IntoIterator<Item = Endpoint>,
127	9	mut connections_per_endpoint: usize,
128	9	mut max_concurrent_requests: usize,
129	9	retry: Retry,
130	9	jitter_fn: retry::JitterFn,
131	9	) -> Self {
132	9	let (worker_tx, worker_rx) = mpsc::channel(WORKER_BACKLOG);
133		// The connection messages always come from sync contexts (e.g. drop)
134		// and therefore, we'd end up spawning for them if this was bounded
135		// which defeats the object since there would be no backpressure
136		// applied. Therefore it makes sense for this to be unbounded.
137	9	let (connection_tx, connection_rx) = mpsc::unbounded_channel();
138	9	let endpoints = endpoints
139	9	.into_iter()
140	9	.map(\|endpoint\| (0, endpoint))
141	9	.collect();
142
143	9	if max_concurrent_requests == 0 {
144	6	max_concurrent_requests = Semaphore::MAX_PERMITS;
145	6	} else {
146	3	max_concurrent_requests = max_concurrent_requests.min(Semaphore::MAX_PERMITS);
147	3	}
148	9	if connections_per_endpoint == 0 {
149	6	connections_per_endpoint = 1;
150	6	}3
151	9	let worker = ConnectionManagerWorker {
152	9	endpoints,
153	9	available_connections: Arc::new(Semaphore::new(max_concurrent_requests)),
154	9	connection_tx,
155	9	connecting_channels: FuturesUnordered::new(),
156	9	available_channels: VecDeque::new(),
157	9	waiting_connections: VecDeque::new(),
158	9	retrier: Retrier::new(
159	9	Arc::new(\|duration\| Box::pin0 ( tokio::time::sleep0 ( duration0 ))),
160	9	jitter_fn,
161	9	retry,
162		),
163		};
164	9	background_spawn!("connection_manager_worker_spawn", async move {8
165	8	worker
166	8	.service_requests(connections_per_endpoint, worker_rx, connection_rx)
167	8	.await;
168	0	});
169	9	Self { worker_tx }
170	9	}
171
172		/// Get a Connection that can be used as a `tonic::Channel`, except it
173		/// performs some additional counting to reconnect on error and restrict
174		/// the number of concurrent connections.
175	220	pub async fn connection(&self, reason: String) -> Result<Connection, Error> {
176	220	let (tx, rx) = oneshot::channel();
177	220	self.worker_tx
178	220	.send((reason, tx))
179	220	.await
180	220	.map_err(\|err\| make_err!0 ( Code::Unavailable0 , "Requesting a new connection: {err:?}")) ?0 ;
181	220	rx.await
182	212	.map_err(\|err\| make_err!0 ( Code::Unavailable0 , "Waiting for a new connection: {err:?}"))
183	212	}
184		}
185
186		impl ConnectionManagerWorker {
187	8	async fn service_requests(
188	8	mut self,
189	8	connections_per_endpoint: usize,
190	8	mut worker_rx: mpsc::Receiver<(String, oneshot::Sender<Connection>)>,
191	8	mut connection_rx: mpsc::UnboundedReceiver<ConnectionRequest>,
192	8	) {
193		// Make the initial set of connections, connection failures will be
194		// handled in the same way as future transport failures, so no need to
195		// do anything special.
196	8	for endpoint_index in 0..self.endpoints.len() {
197	12	for _ in 0..connections_per_endpoint8 {
198	12	self.connect_endpoint(endpoint_index, None);
199	12	}
200		}
201
202		// The main worker loop, when select resolves one of its arms the other
203		// ones are cancelled, therefore it's important that they maintain no
204		// state while `await`-ing. This is enforced through the use of
205		// non-async functions to do all of the work.
206		loop {
207	454	tokio::select! {
208	454	request220 = worker_rx.recv() => {
209	220	let Some((reason, request)) = request else {
210		// The ConnectionManager was dropped, shut down the
211		// worker.
212	0	break;
213		};
214	220	self.handle_worker(reason, request);
215		}
216	454	maybe_request214 = connection_rx.recv() => {
217	214	if let Some(request) = maybe_request {
218	214	self.handle_connection(request);
219	214	}0
220		}
221	454	maybe_connection_result12 = self.connect_next() => {
222	12	if let Some(connection_result) = maybe_connection_result {
223	12	self.handle_connected(connection_result);
224	12	}0
225		}
226		}
227		}
228	0	}
229
230	454	async fn connect_next(&mut self) -> Option<IndexedChannel> {308
231	308	if self.connecting_channels.is_empty() {
232		// Make this Future never resolve, we will get cancelled by the
233		// select if there's some change in state to `self` and can re-enter
234		// and evaluate `connecting_channels` again.
235	287	futures::future::pending::<()>().await;
236	21	}
237	21	self.connecting_channels.next().await
238	12	}
239
240		// This must never be made async otherwise the select may cancel it.
241	12	fn handle_connected(&mut self, connection_result: IndexedChannel) {
242	12	match connection_result {
243	12	Ok(established_channel) => {
244	12	self.available_channels.push_back(established_channel);
245	12	self.maybe_available_connection();
246	12	}
247		// When the retrier runs out of attempts start again from the
248		// beginning of the retry period. Never want to be in a
249		// situation where we give up on an Endpoint forever.
250	0	Err((identifier, _)) => {
251	0	self.connect_endpoint(identifier.endpoint_index, Some(identifier.connection_index));
252	0	}
253		}
254	12	}
255
256	12	fn connect_endpoint(&mut self, endpoint_index: usize, connection_index: Option<usize>) {
257	12	let Some((current_connection_index, endpoint)) = self.endpoints.get_mut(endpoint_index)
258		else {
259		// Unknown endpoint, this should never happen.
260	0	error!(?endpoint_index, "Connection to unknown endpoint requested");
261	0	return;
262		};
263	12	let is_backoff = connection_index.is_some();
264	12	let connection_index = connection_index.unwrap_or_else(\|\| {
265	12	*current_connection_index += 1;
266	12	*current_connection_index
267	12	});
268	12	if is_backoff {
269	0	warn!(
270		?connection_index,
271	0	endpoint = ?endpoint.uri(),
272		"Connection failed, reconnecting"
273		);
274		} else {
275	12	info!(
276		?connection_index,
277	12	endpoint = ?endpoint.uri(),
278		"Creating new connection"
279		);
280		}
281	12	let identifier = ChannelIdentifier {
282	12	endpoint_index,
283	12	connection_index,
284	12	};
285	12	let connection_stream = unfold(endpoint.clone(), move \|endpoint\| async move {
286	12	let result = endpoint.connect().await.map_err(\|err\| {0
287	0	make_err!(
288	0	Code::Unavailable,
289		"Failed to connect to {:?}: {err:?}",
290	0	endpoint.uri()
291		)
292	0	});
293	12	Some((
294	12	result.map_or_else(RetryResult::Retry, RetryResult::Ok),
295	12	endpoint,
296	12	))
297	24	});
298	12	let retrier = self.retrier.clone();
299	12	self.connecting_channels.push(Box::pin(async move {
300	12	if is_backoff {
301		// Just in case the retry config is 0, then we need to
302		// introduce some delay so we aren't in a hard loop.
303	0	tokio::time::sleep(Duration::from_secs(1)).await;
304	12	}
305	12	retrier.retry(connection_stream).await.map_or_else(
306	0	\|err\| Err((identifier, err)),
307	12	\|channel\| {
308	12	Ok(EstablishedChannel {
309	12	channel,
310	12	identifier,
311	12	})
312	12	},
313		)
314	12	}));
315	12	}
316
317		// This must never be made async otherwise the select may cancel it.
318	220	fn handle_worker(&mut self, reason: String, tx: oneshot::Sender<Connection>) {
319	220	let maybe_permit = self.available_connections.clone().try_acquire_owned().ok();
320	220	if let Some( permit219 ) = maybe_permit
321	219	&& let Some( channel140 ) = self.available_channels.pop_front()
322		{
323	140	debug!(reason, "ConnectionManager: request running");
324	140	self.provide_channel(channel, tx, permit);
325		} else {
326	80	debug!(
327	80	available_permits = self.available_connections.available_permits(),
328	80	available_channels = self.available_channels.len(),
329	80	waiting_connections = self.waiting_connections.len(),
330		reason,
331		"ConnectionManager: no connection available, request queued",
332		);
333	80	self.waiting_connections.push_back((reason, tx));
334		}
335	220	}
336
337	220	fn provide_channel(
338	220	&self,
339	220	channel: EstablishedChannel,
340	220	tx: oneshot::Sender<Connection>,
341	220	permit: OwnedSemaphorePermit,
342	220	) {
343	220	drop(tx.send(Connection {
344	220	tx: self.connection_tx.clone(),
345	220	pending_channel: Some(channel.channel.clone()),
346	220	channel,
347	220	_permit: permit,
348	220	}));
349	220	}
350
351	226	fn maybe_available_connection(&mut self) {
352	306	while !self.waiting_connections.is_empty() && !self.available_channels.is_empty()89 {
353	80	let Some(permit) = self.available_connections.clone().try_acquire_owned().ok() else {
354	0	break;
355		};
356	80	let Some(channel) = self.available_channels.pop_front() else {
357	0	drop(permit);
358	0	break;
359		};
360	80	let Some((reason, tx)) = self.waiting_connections.pop_front() else {
361	0	self.available_channels.push_front(channel);
362	0	drop(permit);
363	0	break;
364		};
365	80	debug!(reason, "ConnectionManager: channel available, running");
366	80	self.provide_channel(channel, tx, permit);
367		}
368	226	}
369
370		// This must never be made async otherwise the select may cancel it.
371	214	fn handle_connection(&mut self, request: ConnectionRequest) {
372	214	match request {
373	209	ConnectionRequest::Dropped(maybe_channel) => {
374	209	if let Some(channel) = maybe_channel {
375	209	self.available_channels.push_back(channel);
376	209	}0
377	209	self.maybe_available_connection();
378		}
379	5	ConnectionRequest::Connected(channel) => {
380	5	self.available_channels.push_back(channel);
381	5	self.maybe_available_connection();
382	5	}
383		// Handle a transport error on a connection by making it unavailable
384		// for use and establishing a new connection to the endpoint.
385	0	ConnectionRequest::Error((identifier, was_pending)) => {
386	0	let should_reconnect = if was_pending {
387	0	true
388		} else {
389	0	let original_length = self.available_channels.len();
390	0	self.available_channels
391	0	.retain(\|channel\| channel.identifier != identifier);
392		// Only reconnect if it wasn't already disconnected.
393	0	original_length != self.available_channels.len()
394		};
395	0	if should_reconnect {
396	0	self.connect_endpoint(identifier.endpoint_index, None);
397	0	}
398		}
399		}
400	214	}
401		}
402
403		/// An instance of this is obtained for every communication with the gGRPC
404		/// service. This handles the permit for limiting concurrency, and also
405		/// re-connecting the underlying channel on error. It depends on users
406		/// reporting all errors.
407		/// NOTE: This should never be cloneable because its lifetime is linked to the
408		/// semaphore permit it carries — `_permit` is released exactly once,
409		/// when the `Connection` drops.
410		#[derive(Debug)]
411		pub struct Connection {
412		/// Communication with `ConnectionManagerWorker` to inform about transport
413		/// errors and when the Connection is dropped.
414		tx: mpsc::UnboundedSender<ConnectionRequest>,
415		/// If set, the Channel that will be returned to the worker when connection
416		/// completes (success or failure) or when the Connection is dropped if that
417		/// happens before connection completes.
418		pending_channel: Option<Channel>,
419		/// The identifier to send to `tx`.
420		channel: EstablishedChannel,
421		_permit: OwnedSemaphorePermit,
422		}
423
424		impl Drop for Connection {
425	220	fn drop(&mut self) {
426	220	let pending_channel = self
427	220	.pending_channel
428	220	.take()
429	220	.map(\|channel\| EstablishedChannel {
430	215	channel,
431	215	identifier: self.channel.identifier,
432	215	});
433	220	drop(self.tx.send(ConnectionRequest::Dropped(pending_channel)));
434	220	}
435		}
436
437		/// A wrapper around the `channel::ResponseFuture` that forwards errors to the `tx`.
438		pub struct ResponseFuture {
439		/// The wrapped future that actually does the work.
440		inner: channel::ResponseFuture,
441		/// Communication with `ConnectionManagerWorker` to inform about transport
442		/// errors.
443		connection_tx: mpsc::UnboundedSender<ConnectionRequest>,
444		/// The identifier to send to `connection_tx` on a transport error.
445		identifier: ChannelIdentifier,
446		}
447
448		impl core::fmt::Debug for ResponseFuture {
449	0	fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
450	0	f.debug_struct("ResponseFuture")
451	0	.field("inner", &self.inner)
452	0	.field("connection_tx", &self.connection_tx)
453	0	.field("identifier", &self.identifier)
454	0	.finish()
455	0	}
456		}
457
458		/// This is mostly copied from `tonic::transport::channel` except it wraps it
459		/// to allow messaging about connection success and failure.
460		impl tonic::codegen::Service<tonic::codegen::http::Request<tonic::body::Body>> for Connection {
461		type Response = tonic::codegen::http::Response<tonic::body::Body>;
462		type Error = tonic::transport::Error;
463		type Future = ResponseFuture;
464
465	5	fn poll_ready(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
466	5	let result = self.channel.channel.poll_ready(cx);
467	5	if let Poll::Ready(result) = &result {
468	5	match result {
469		Ok(()) => {
470	5	if let Some(pending_channel) = self.pending_channel.take() {
471	5	drop(
472	5	self.tx
473	5	.send(ConnectionRequest::Connected(EstablishedChannel {
474	5	channel: pending_channel,
475	5	identifier: self.channel.identifier,
476	5	})),
477	5	);
478	5	}0
479		}
480	0	Err(err) => {
481	0	debug!(?err, "Error while creating connection on channel");
482	0	drop(self.tx.send(ConnectionRequest::Error((
483	0	self.channel.identifier,
484	0	self.pending_channel.take().is_some(),
485	0	))));
486		}
487		}
488	0	}
489	5	result
490	5	}
491
492	5	fn call(&mut self, request: tonic::codegen::http::Request<tonic::body::Body>) -> Self::Future {
493	5	ResponseFuture {
494	5	inner: self.channel.channel.call(request),
495	5	connection_tx: self.tx.clone(),
496	5	identifier: self.channel.identifier,
497	5	}
498	5	}
499		}
500
501		/// This is mostly copied from `tonic::transport::channel` except it wraps it
502		/// to allow messaging about connection failure.
503		impl Future for ResponseFuture {
504		type Output =
505		Result<tonic::codegen::http::Response<tonic::body::Body>, tonic::transport::Error>;
506
507	15	fn poll(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
508	15	let result = Pin::new(&mut self.inner).poll(cx);
509	5	if let Poll::Ready(Err(_)) = &result {
510	0	drop(
511	0	self.connection_tx
512	0	.send(ConnectionRequest::Error((self.identifier, false))),
513	0	);
514	15	}
515	15	result
516	15	}
517		}