diff --git a/docs/09_sending_activities.md b/docs/09_sending_activities.md index 215be5e..d5efe3a 100644 --- a/docs/09_sending_activities.md +++ b/docs/09_sending_activities.md @@ -4,6 +4,46 @@ To send an activity we need to initialize our previously defined struct, and pic ``` # use activitypub_federation::config::FederationConfig; +# use activitypub_federation::activity_queue::queue_activity; +# use activitypub_federation::http_signatures::generate_actor_keypair; +# use activitypub_federation::traits::Actor; +# use activitypub_federation::fetch::object_id::ObjectId; +# use activitypub_federation::traits::tests::{DB_USER, DbConnection, Follow}; +# tokio::runtime::Runtime::new().unwrap().block_on(async { +# let db_connection = DbConnection; +# let config = FederationConfig::builder() +# .domain("example.com") +# .app_data(db_connection) +# .build().await?; +# let data = config.to_request_data(); +# let sender = DB_USER.clone(); +# let recipient = DB_USER.clone(); +let activity = Follow { + actor: ObjectId::parse("https://lemmy.ml/u/nutomic")?, + object: recipient.federation_id.clone().into(), + kind: Default::default(), + id: "https://lemmy.ml/activities/321".try_into()? +}; +let inboxes = vec![recipient.shared_inbox_or_inbox()]; + +queue_activity(&activity, &sender, inboxes, &data).await?; +# Ok::<(), anyhow::Error>(()) +# }).unwrap() +``` + +The list of inboxes gets deduplicated (important for shared inbox). All inboxes on the local domain and those which fail the [crate::config::UrlVerifier] check are excluded from delivery. For each remaining inbox a background tasks is created. It signs the HTTP header with the given private key. Finally the activity is delivered to the inbox. + +It is possible that delivery fails because the target instance is temporarily unreachable. In this case the task is scheduled for retry after a certain waiting time. For each task delivery is retried up to 3 times after the initial attempt. The retry intervals are as follows: + +- one minute, in case of service restart +- one hour, in case of instance maintenance +- 2.5 days, in case of major incident with rebuild from backup + +In case [crate::config::FederationConfigBuilder::debug] is enabled, no background thread is used but activities are sent directly on the foreground. This makes it easier to catch delivery errors and avoids complicated steps to await delivery in tests. + +In some cases you may want to bypass the builtin activity queue, and implement your own. For example to specify different retry intervals, or to persist retries across application restarts. You can do it with the following code: +```rust +# use activitypub_federation::config::FederationConfig; # use activitypub_federation::activity_sending::SendActivityTask; # use activitypub_federation::http_signatures::generate_actor_keypair; # use activitypub_federation::traits::Actor; @@ -28,23 +68,8 @@ let inboxes = vec![recipient.shared_inbox_or_inbox()]; let sends = SendActivityTask::prepare(&activity, &sender, inboxes, &data).await?; for send in sends { - send.sign_and_send(&data).await?; +send.sign_and_send(&data).await?; } # Ok::<(), anyhow::Error>(()) # }).unwrap() -``` - -The list of inboxes gets deduplicated (important for shared inbox). All inboxes on the local -domain and those which fail the [crate::config::UrlVerifier] check are excluded from delivery. -For each remaining inbox a background tasks is created. It signs the HTTP header with the given -private key. Finally the activity is delivered to the inbox. - -It is possible that delivery fails because the target instance is temporarily unreachable. In -this case the task is scheduled for retry after a certain waiting time. For each task delivery -is retried up to 3 times after the initial attempt. The retry intervals are as follows: - -- one minute, in case of service restart -- one hour, in case of instance maintenance -- 2.5 days, in case of major incident with rebuild from backup - -In case [crate::config::FederationConfigBuilder::debug] is enabled, no background thread is used but activities are sent directly on the foreground. This makes it easier to catch delivery errors and avoids complicated steps to await delivery in tests. +``` \ No newline at end of file diff --git a/examples/local_federation/activities/follow.rs b/examples/local_federation/activities/follow.rs index 865a618..02879d4 100644 --- a/examples/local_federation/activities/follow.rs +++ b/examples/local_federation/activities/follow.rs @@ -67,7 +67,7 @@ impl ActivityHandler for Follow { let id = generate_object_id(data.domain())?; let accept = Accept::new(local_user.ap_id.clone(), self, id.clone()); local_user - .send(accept, vec![follower.shared_inbox_or_inbox()], data) + .send(accept, vec![follower.shared_inbox_or_inbox()], false, data) .await?; Ok(()) } diff --git a/examples/local_federation/objects/person.rs b/examples/local_federation/objects/person.rs index 2c47fcd..0ae402f 100644 --- a/examples/local_federation/objects/person.rs +++ b/examples/local_federation/objects/person.rs @@ -6,6 +6,7 @@ use crate::{ utils::generate_object_id, }; use activitypub_federation::{ + activity_queue::queue_activity, activity_sending::SendActivityTask, config::Data, fetch::{object_id::ObjectId, webfinger::webfinger_resolve_actor}, @@ -85,7 +86,7 @@ impl DbUser { let other: DbUser = webfinger_resolve_actor(other, data).await?; let id = generate_object_id(data.domain())?; let follow = Follow::new(self.ap_id.clone(), other.ap_id.clone(), id.clone()); - self.send(follow, vec![other.shared_inbox_or_inbox()], data) + self.send(follow, vec![other.shared_inbox_or_inbox()], false, data) .await?; Ok(()) } @@ -98,7 +99,7 @@ impl DbUser { let user: DbUser = ObjectId::from(f).dereference(data).await?; inboxes.push(user.shared_inbox_or_inbox()); } - self.send(create, inboxes, data).await?; + self.send(create, inboxes, true, data).await?; Ok(()) } @@ -106,6 +107,7 @@ impl DbUser { &self, activity: Activity, recipients: Vec, + use_queue: bool, data: &Data, ) -> Result<(), Error> where @@ -113,9 +115,14 @@ impl DbUser { ::Error: From + From, { let activity = WithContext::new_default(activity); - let sends = SendActivityTask::prepare(&activity, self, recipients, data).await?; - for send in sends { - send.sign_and_send(data).await?; + // Send through queue in some cases and bypass it in others to test both code paths + if use_queue { + queue_activity(&activity, self, recipients, data).await?; + } else { + let sends = SendActivityTask::prepare(&activity, self, recipients, data).await?; + for send in sends { + send.sign_and_send(data).await?; + } } Ok(()) } diff --git a/src/activity_queue.rs b/src/activity_queue.rs new file mode 100644 index 0000000..abb5800 --- /dev/null +++ b/src/activity_queue.rs @@ -0,0 +1,520 @@ +//! Queue for signing and sending outgoing activities with retry +//! +#![doc = include_str!("../docs/09_sending_activities.md")] + +use crate::{ + activity_sending::{build_tasks, SendActivityTask}, + config::Data, + error::Error, + traits::{ActivityHandler, Actor}, +}; + +use futures_core::Future; + +use reqwest_middleware::ClientWithMiddleware; +use serde::Serialize; +use std::{ + fmt::{Debug, Display}, + sync::{ + atomic::{AtomicUsize, Ordering}, + Arc, + }, + time::Duration, +}; +use tokio::{ + sync::mpsc::{unbounded_channel, UnboundedSender}, + task::{JoinHandle, JoinSet}, +}; +use tracing::{info, warn}; +use url::Url; + +/// Send a new activity to the given inboxes with automatic retry on failure. Alternatively you +/// can implement your own queue and then send activities using [[crate::activity_sending::SendActivityTask]]. +/// +/// - `activity`: The activity to be sent, gets converted to json +/// - `private_key`: Private key belonging to the actor who sends the activity, for signing HTTP +/// signature. Generated with [crate::http_signatures::generate_actor_keypair]. +/// - `inboxes`: List of remote actor inboxes that should receive the activity. Ignores local actor +/// inboxes. Should be built by calling [crate::traits::Actor::shared_inbox_or_inbox] +/// for each target actor. +pub async fn queue_activity( + activity: &Activity, + actor: &ActorType, + inboxes: Vec, + data: &Data, +) -> Result<(), Error> +where + Activity: ActivityHandler + Serialize + Debug, + Datatype: Clone, + ActorType: Actor, +{ + let config = &data.config; + let tasks = build_tasks(activity, actor, inboxes, data).await?; + + for task in tasks { + // Don't use the activity queue if this is in debug mode, send and wait directly + if config.debug { + if let Err(err) = sign_and_send( + &task, + &config.client, + config.request_timeout, + Default::default(), + ) + .await + { + warn!("{err}"); + } + } else { + // This field is only optional to make builder work, its always present at this point + let activity_queue = config + .activity_queue + .as_ref() + .expect("Config has activity queue"); + activity_queue.queue(task).await?; + let stats = activity_queue.get_stats(); + let running = stats.running.load(Ordering::Relaxed); + if running == config.queue_worker_count && config.queue_worker_count != 0 { + warn!("Reached max number of send activity workers ({}). Consider increasing worker count to avoid federation delays", config.queue_worker_count); + warn!("{:?}", stats); + } else { + info!("{:?}", stats); + } + } + } + Ok(()) +} + +async fn sign_and_send( + task: &SendActivityTask, + client: &ClientWithMiddleware, + timeout: Duration, + retry_strategy: RetryStrategy, +) -> Result<(), Error> { + retry( + || task.sign_and_send_internal(client, timeout), + retry_strategy, + ) + .await +} + +/// A simple activity queue which spawns tokio workers to send out requests +/// When creating a queue, it will spawn a task per worker thread +/// Uses an unbounded mpsc queue for communication (i.e, all messages are in memory) +pub(crate) struct ActivityQueue { + // Stats shared between the queue and workers + stats: Arc, + sender: UnboundedSender, + sender_task: JoinHandle<()>, + retry_sender_task: JoinHandle<()>, +} + +/// Simple stat counter to show where we're up to with sending messages +/// This is a lock-free way to share things between tasks +/// When reading these values it's possible (but extremely unlikely) to get stale data if a worker task is in the middle of transitioning +#[derive(Default)] +pub(crate) struct Stats { + pending: AtomicUsize, + running: AtomicUsize, + retries: AtomicUsize, + dead_last_hour: AtomicUsize, + completed_last_hour: AtomicUsize, +} + +impl Debug for Stats { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + write!( + f, + "Activity queue stats: pending: {}, running: {}, retries: {}, dead: {}, complete: {}", + self.pending.load(Ordering::Relaxed), + self.running.load(Ordering::Relaxed), + self.retries.load(Ordering::Relaxed), + self.dead_last_hour.load(Ordering::Relaxed), + self.completed_last_hour.load(Ordering::Relaxed) + ) + } +} + +#[derive(Clone, Copy, Default)] +struct RetryStrategy { + /// Amount of time in seconds to back off + backoff: usize, + /// Amount of times to retry + retries: usize, + /// If this particular request has already been retried, you can add an offset here to increment the count to start + offset: usize, + /// Number of seconds to sleep before trying + initial_sleep: usize, +} + +/// A tokio spawned worker which is responsible for submitting requests to federated servers +/// This will retry up to one time with the same signature, and if it fails, will move it to the retry queue. +/// We need to retry activity sending in case the target instances is temporarily unreachable. +/// In this case, the task is stored and resent when the instance is hopefully back up. This +/// list shows the retry intervals, and which events of the target instance can be covered: +/// - 60s (one minute, service restart) -- happens in the worker w/ same signature +/// - 60min (one hour, instance maintenance) --- happens in the retry worker +/// - 60h (2.5 days, major incident with rebuild from backup) --- happens in the retry worker +async fn worker( + client: ClientWithMiddleware, + timeout: Duration, + message: SendActivityTask, + retry_queue: UnboundedSender, + stats: Arc, + strategy: RetryStrategy, +) { + stats.pending.fetch_sub(1, Ordering::Relaxed); + stats.running.fetch_add(1, Ordering::Relaxed); + + let outcome = sign_and_send(&message, &client, timeout, strategy).await; + + // "Running" has finished, check the outcome + stats.running.fetch_sub(1, Ordering::Relaxed); + + match outcome { + Ok(_) => { + stats.completed_last_hour.fetch_add(1, Ordering::Relaxed); + } + Err(_err) => { + stats.retries.fetch_add(1, Ordering::Relaxed); + warn!( + "Sending activity {} to {} to the retry queue to be tried again later", + message.activity_id, message.inbox + ); + // Send to the retry queue. Ignoring whether it succeeds or not + retry_queue.send(message).ok(); + } + } +} + +async fn retry_worker( + client: ClientWithMiddleware, + timeout: Duration, + message: SendActivityTask, + stats: Arc, + strategy: RetryStrategy, +) { + // Because the times are pretty extravagant between retries, we have to re-sign each time + let outcome = retry( + || { + sign_and_send( + &message, + &client, + timeout, + RetryStrategy { + backoff: 0, + retries: 0, + offset: 0, + initial_sleep: 0, + }, + ) + }, + strategy, + ) + .await; + + stats.retries.fetch_sub(1, Ordering::Relaxed); + + match outcome { + Ok(_) => { + stats.completed_last_hour.fetch_add(1, Ordering::Relaxed); + } + Err(_err) => { + stats.dead_last_hour.fetch_add(1, Ordering::Relaxed); + } + } +} + +impl ActivityQueue { + fn new( + client: ClientWithMiddleware, + worker_count: usize, + retry_count: usize, + timeout: Duration, + backoff: usize, // This should be 60 seconds by default or 1 second in tests + ) -> Self { + let stats: Arc = Default::default(); + + // This task clears the dead/completed stats every hour + let hour_stats = stats.clone(); + tokio::spawn(async move { + let duration = Duration::from_secs(3600); + loop { + tokio::time::sleep(duration).await; + hour_stats.completed_last_hour.store(0, Ordering::Relaxed); + hour_stats.dead_last_hour.store(0, Ordering::Relaxed); + } + }); + + let (retry_sender, mut retry_receiver) = unbounded_channel(); + let retry_stats = stats.clone(); + let retry_client = client.clone(); + + // The "fast path" retry + // The backoff should be < 5 mins for this to work otherwise signatures may expire + // This strategy is the one that is used with the *same* signature + let strategy = RetryStrategy { + backoff, + retries: 1, + offset: 0, + initial_sleep: 0, + }; + + // The "retry path" strategy + // After the fast path fails, a task will sleep up to backoff ^ 2 and then retry again + let retry_strategy = RetryStrategy { + backoff, + retries: 3, + offset: 2, + initial_sleep: backoff.pow(2), // wait 60 mins before even trying + }; + + let retry_sender_task = tokio::spawn(async move { + let mut join_set = JoinSet::new(); + + while let Some(message) = retry_receiver.recv().await { + let retry_task = retry_worker( + retry_client.clone(), + timeout, + message, + retry_stats.clone(), + retry_strategy, + ); + + if retry_count > 0 { + // If we're over the limit of retries, wait for them to finish before spawning + while join_set.len() >= retry_count { + join_set.join_next().await; + } + + join_set.spawn(retry_task); + } else { + // If the retry worker count is `0` then just spawn and don't use the join_set + tokio::spawn(retry_task); + } + } + + while !join_set.is_empty() { + join_set.join_next().await; + } + }); + + let (sender, mut receiver) = unbounded_channel(); + + let sender_stats = stats.clone(); + + let sender_task = tokio::spawn(async move { + let mut join_set = JoinSet::new(); + + while let Some(message) = receiver.recv().await { + let task = worker( + client.clone(), + timeout, + message, + retry_sender.clone(), + sender_stats.clone(), + strategy, + ); + + if worker_count > 0 { + // If we're over the limit of workers, wait for them to finish before spawning + while join_set.len() >= worker_count { + join_set.join_next().await; + } + + join_set.spawn(task); + } else { + // If the worker count is `0` then just spawn and don't use the join_set + tokio::spawn(task); + } + } + + drop(retry_sender); + + while !join_set.is_empty() { + join_set.join_next().await; + } + }); + + Self { + stats, + sender, + sender_task, + retry_sender_task, + } + } + + async fn queue(&self, message: SendActivityTask) -> Result<(), Error> { + self.stats.pending.fetch_add(1, Ordering::Relaxed); + self.sender + .send(message) + .map_err(|e| Error::ActivityQueueError(e.0.activity_id))?; + + Ok(()) + } + + fn get_stats(&self) -> &Stats { + &self.stats + } + + #[allow(unused)] + // Drops all the senders and shuts down the workers + pub(crate) async fn shutdown(self, wait_for_retries: bool) -> Result, Error> { + drop(self.sender); + + self.sender_task.await?; + + if wait_for_retries { + self.retry_sender_task.await?; + } + + Ok(self.stats) + } +} + +/// Creates an activity queue using tokio spawned tasks +/// Note: requires a tokio runtime +pub(crate) fn create_activity_queue( + client: ClientWithMiddleware, + worker_count: usize, + retry_count: usize, + request_timeout: Duration, +) -> ActivityQueue { + ActivityQueue::new(client, worker_count, retry_count, request_timeout, 60) +} + +/// Retries a future action factory function up to `amount` times with an exponential backoff timer between tries +async fn retry>, A: FnMut() -> F>( + mut action: A, + strategy: RetryStrategy, +) -> Result { + let mut count = strategy.offset; + + // Do an initial sleep if it's called for + if strategy.initial_sleep > 0 { + let sleep_dur = Duration::from_secs(strategy.initial_sleep as u64); + tokio::time::sleep(sleep_dur).await; + } + + loop { + match action().await { + Ok(val) => return Ok(val), + Err(err) => { + if count < strategy.retries { + count += 1; + + let sleep_amt = strategy.backoff.pow(count as u32) as u64; + let sleep_dur = Duration::from_secs(sleep_amt); + warn!("{err:?}. Sleeping for {sleep_dur:?} and trying again"); + tokio::time::sleep(sleep_dur).await; + continue; + } else { + return Err(err); + } + } + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::http_signatures::generate_actor_keypair; + use axum::extract::State; + use bytes::Bytes; + use http::{HeaderMap, StatusCode}; + use std::time::Instant; + use tracing::debug; + + // This will periodically send back internal errors to test the retry + async fn dodgy_handler( + State(state): State>, + headers: HeaderMap, + body: Bytes, + ) -> Result<(), StatusCode> { + debug!("Headers:{:?}", headers); + debug!("Body len:{}", body.len()); + + if state.fetch_add(1, Ordering::Relaxed) % 20 == 0 { + return Err(StatusCode::INTERNAL_SERVER_ERROR); + } + Ok(()) + } + + async fn test_server() { + use axum::{routing::post, Router}; + + // We should break every now and then ;) + let state = Arc::new(AtomicUsize::new(0)); + + let app = Router::new() + .route("/", post(dodgy_handler)) + .with_state(state); + + axum::Server::bind(&"0.0.0.0:8002".parse().unwrap()) + .serve(app.into_make_service()) + .await + .unwrap(); + } + + #[tokio::test(flavor = "multi_thread")] + // Queues 100 messages and then asserts that the worker runs them + async fn test_activity_queue_workers() { + let num_workers = 64; + let num_messages: usize = 100; + + tokio::spawn(test_server()); + + /* + // uncomment for debug logs & stats + use tracing::log::LevelFilter; + + env_logger::builder() + .filter_level(LevelFilter::Warn) + .filter_module("activitypub_federation", LevelFilter::Info) + .format_timestamp(None) + .init(); + + */ + + let activity_queue = ActivityQueue::new( + reqwest::Client::default().into(), + num_workers, + num_workers, + Duration::from_secs(10), + 1, + ); + + let keypair = generate_actor_keypair().unwrap(); + + let message = SendActivityTask { + actor_id: "http://localhost:8002".parse().unwrap(), + activity_id: "http://localhost:8002/activity".parse().unwrap(), + activity: "{}".into(), + inbox: "http://localhost:8002".parse().unwrap(), + private_key: keypair.private_key().unwrap(), + http_signature_compat: true, + }; + + let start = Instant::now(); + + for _ in 0..num_messages { + activity_queue.queue(message.clone()).await.unwrap(); + } + + info!("Queue Sent: {:?}", start.elapsed()); + + let stats = activity_queue.shutdown(true).await.unwrap(); + + info!( + "Queue Finished. Num msgs: {}, Time {:?}, msg/s: {:0.0}", + num_messages, + start.elapsed(), + num_messages as f64 / start.elapsed().as_secs_f64() + ); + + assert_eq!( + stats.completed_last_hour.load(Ordering::Relaxed), + num_messages + ); + } +} diff --git a/src/activity_sending.rs b/src/activity_sending.rs index bd236ad..f16ef37 100644 --- a/src/activity_sending.rs +++ b/src/activity_sending.rs @@ -10,98 +10,80 @@ use crate::{ traits::{ActivityHandler, Actor}, FEDERATION_CONTENT_TYPE, }; - use bytes::Bytes; use futures::StreamExt; use httpdate::fmt_http_date; use itertools::Itertools; use openssl::pkey::{PKey, Private}; use reqwest::header::{HeaderMap, HeaderName, HeaderValue}; +use reqwest_middleware::ClientWithMiddleware; use serde::Serialize; use std::{ self, fmt::{Debug, Display}, - time::SystemTime, + time::{Duration, SystemTime}, }; use tracing::debug; use url::Url; #[derive(Clone, Debug)] -/// all info needed to send one activity to one inbox -pub struct SendActivityTask<'a> { - actor_id: &'a Url, - activity_id: &'a Url, - activity: Bytes, - inbox: Url, - private_key: PKey, - http_signature_compat: bool, +/// All info needed to sign and send one activity to one inbox. You should generally use +/// [[crate::activity_queue::queue_activity]] unless you want implement your own queue. +pub struct SendActivityTask { + pub(crate) actor_id: Url, + pub(crate) activity_id: Url, + pub(crate) activity: Bytes, + pub(crate) inbox: Url, + pub(crate) private_key: PKey, + pub(crate) http_signature_compat: bool, } -impl Display for SendActivityTask<'_> { + +impl Display for SendActivityTask { fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { write!(f, "{} to {}", self.activity_id, self.inbox) } } -impl SendActivityTask<'_> { - /// prepare an activity for sending +impl SendActivityTask { + /// Prepare an activity for sending /// /// - `activity`: The activity to be sent, gets converted to json /// - `inboxes`: List of remote actor inboxes that should receive the activity. Ignores local actor /// inboxes. Should be built by calling [crate::traits::Actor::shared_inbox_or_inbox] /// for each target actor. - pub async fn prepare<'a, Activity, Datatype, ActorType>( - activity: &'a Activity, + pub async fn prepare( + activity: &Activity, actor: &ActorType, inboxes: Vec, data: &Data, - ) -> Result>, Error> + ) -> Result, Error> where Activity: ActivityHandler + Serialize + Debug, Datatype: Clone, ActorType: Actor, { - let config = &data.config; - let actor_id = activity.actor(); - let activity_id = activity.id(); - let activity_serialized: Bytes = serde_json::to_vec(&activity) - .map_err(|e| Error::SerializeOutgoingActivity(e, format!("{:?}", activity)))? - .into(); - let private_key = get_pkey_cached(data, actor).await?; - - Ok(futures::stream::iter( - inboxes - .into_iter() - .unique() - .filter(|i| !config.is_local_url(i)), - ) - .filter_map(|inbox| async { - if let Err(err) = config.verify_url_valid(&inbox).await { - debug!("inbox url invalid, skipping: {inbox}: {err}"); - return None; - }; - Some(SendActivityTask { - actor_id, - activity_id, - inbox, - activity: activity_serialized.clone(), - private_key: private_key.clone(), - http_signature_compat: config.http_signature_compat, - }) - }) - .collect() - .await) + build_tasks(activity, actor, inboxes, data).await } /// convert a sendactivitydata to a request, signing and sending it pub async fn sign_and_send(&self, data: &Data) -> Result<(), Error> { - let client = &data.config.client; + self.sign_and_send_internal(&data.config.client, data.config.request_timeout) + .await + } + + pub(crate) async fn sign_and_send_internal( + &self, + client: &ClientWithMiddleware, + timeout: Duration, + ) -> Result<(), Error> { + debug!("Sending {} to {}", self.activity_id, self.inbox,); let request_builder = client .post(self.inbox.to_string()) - .timeout(data.config.request_timeout) + .timeout(timeout) .headers(generate_request_headers(&self.inbox)); let request = sign_request( request_builder, - self.actor_id, + &self.actor_id, self.activity.clone(), self.private_key.clone(), self.http_signature_compat, @@ -131,7 +113,50 @@ impl SendActivityTask<'_> { } } -async fn get_pkey_cached( +pub(crate) async fn build_tasks<'a, Activity, Datatype, ActorType>( + activity: &'a Activity, + actor: &ActorType, + inboxes: Vec, + data: &Data, +) -> Result, Error> +where + Activity: ActivityHandler + Serialize + Debug, + Datatype: Clone, + ActorType: Actor, +{ + let config = &data.config; + let actor_id = activity.actor(); + let activity_id = activity.id(); + let activity_serialized: Bytes = serde_json::to_vec(activity) + .map_err(|e| Error::SerializeOutgoingActivity(e, format!("{:?}", activity)))? + .into(); + let private_key = get_pkey_cached(data, actor).await?; + + Ok(futures::stream::iter( + inboxes + .into_iter() + .unique() + .filter(|i| !config.is_local_url(i)), + ) + .filter_map(|inbox| async { + if let Err(err) = config.verify_url_valid(&inbox).await { + debug!("inbox url invalid, skipping: {inbox}: {err}"); + return None; + }; + Some(SendActivityTask { + actor_id: actor_id.clone(), + activity_id: activity_id.clone(), + inbox, + activity: activity_serialized.clone(), + private_key: private_key.clone(), + http_signature_compat: config.http_signature_compat, + }) + }) + .collect() + .await) +} + +pub(crate) async fn get_pkey_cached( data: &Data, actor: &ActorType, ) -> Result, Error> @@ -187,7 +212,8 @@ pub(crate) fn generate_request_headers(inbox_url: &Url) -> HeaderMap { #[cfg(test)] mod tests { - use axum::extract::State; + use super::*; + use crate::{config::FederationConfig, http_signatures::generate_actor_keypair}; use bytes::Bytes; use http::StatusCode; use std::{ @@ -196,23 +222,10 @@ mod tests { }; use tracing::info; - use crate::{config::FederationConfig, http_signatures::generate_actor_keypair}; - - use super::*; - - #[allow(unused)] // This will periodically send back internal errors to test the retry - async fn dodgy_handler( - State(state): State>, - headers: HeaderMap, - body: Bytes, - ) -> Result<(), StatusCode> { + async fn dodgy_handler(headers: HeaderMap, body: Bytes) -> Result<(), StatusCode> { debug!("Headers:{:?}", headers); debug!("Body len:{}", body.len()); - - /*if state.fetch_add(1, Ordering::Relaxed) % 20 == 0 { - return Err(StatusCode::INTERNAL_SERVER_ERROR); - }*/ Ok(()) } @@ -253,8 +266,8 @@ mod tests { let keypair = generate_actor_keypair().unwrap(); let message = SendActivityTask { - actor_id: &"http://localhost:8001".parse().unwrap(), - activity_id: &"http://localhost:8001/activity".parse().unwrap(), + actor_id: "http://localhost:8001".parse().unwrap(), + activity_id: "http://localhost:8001/activity".parse().unwrap(), activity: "{}".into(), inbox: "http://localhost:8001".parse().unwrap(), private_key: keypair.private_key().unwrap(), diff --git a/src/config.rs b/src/config.rs index 7573234..b7aea9c 100644 --- a/src/config.rs +++ b/src/config.rs @@ -15,6 +15,7 @@ //! ``` use crate::{ + activity_queue::{create_activity_queue, ActivityQueue}, error::Error, protocol::verification::verify_domains_match, traits::{ActivityHandler, Actor}, @@ -85,6 +86,20 @@ pub struct FederationConfig { setter(custom) )] pub(crate) actor_pkey_cache: Cache>, + /// Queue for sending outgoing activities. Only optional to make builder work, its always + /// present once constructed. + #[builder(setter(skip))] + pub(crate) activity_queue: Option>, + /// When sending with activity queue: Number of tasks that can be in-flight concurrently. + /// Tasks are retried once after a minute, then put into the retry queue. + /// Setting this count to `0` means that there is no limit to concurrency + #[builder(default = "0")] + pub(crate) queue_worker_count: usize, + /// When sending with activity queue: Number of concurrent tasks that are being retried + /// in-flight concurrently. Tasks are retried after an hour, then again in 60 hours. + /// Setting this count to `0` means that there is no limit to concurrency + #[builder(default = "0")] + pub(crate) queue_retry_count: usize, } impl FederationConfig { @@ -197,7 +212,14 @@ impl FederationConfigBuilder { /// queue for outgoing activities, which is stored internally in the config struct. /// Requires a tokio runtime for the background queue. pub async fn build(&mut self) -> Result, FederationConfigBuilderError> { - let config = self.partial_build()?; + let mut config = self.partial_build()?; + let queue = create_activity_queue( + config.client.clone(), + config.queue_worker_count, + config.queue_retry_count, + config.request_timeout, + ); + config.activity_queue = Some(Arc::new(queue)); Ok(config) } } diff --git a/src/error.rs b/src/error.rs index 1383ddb..d2f7c87 100644 --- a/src/error.rs +++ b/src/error.rs @@ -4,6 +4,7 @@ use crate::fetch::webfinger::WebFingerError; use http_signature_normalization_reqwest::SignError; use openssl::error::ErrorStack; use std::string::FromUtf8Error; +use tokio::task::JoinError; use url::Url; /// Error messages returned by this library @@ -60,6 +61,12 @@ pub enum Error { /// Signing errors #[error(transparent)] SignError(#[from] SignError), + /// Failed to queue activity for sending + #[error("Failed to queue activity {0} for sending")] + ActivityQueueError(Url), + /// Stop activity queue + #[error(transparent)] + StopActivityQueue(#[from] JoinError), /// Attempted to fetch object which doesn't have valid ActivityPub Content-Type #[error( "Attempted to fetch object from {0} which doesn't have valid ActivityPub Content-Type" diff --git a/src/http_signatures.rs b/src/http_signatures.rs index 59d1d48..bc27ee5 100644 --- a/src/http_signatures.rs +++ b/src/http_signatures.rs @@ -1,7 +1,7 @@ //! Generating keypairs, creating and verifying signatures //! //! Signature creation and verification is handled internally in the library. See -//! [send_activity](crate::activity_sending::send_activity) and +//! [send_activity](crate::activity_sending::SendActivityTask::sign_and_send) and //! [receive_activity (actix-web)](crate::actix_web::inbox::receive_activity) / //! [receive_activity (axum)](crate::axum::inbox::receive_activity). diff --git a/src/lib.rs b/src/lib.rs index c025d29..0a44fc9 100644 --- a/src/lib.rs +++ b/src/lib.rs @@ -10,6 +10,7 @@ #![doc = include_str!("../docs/10_fetching_objects_with_unknown_type.md")] #![deny(missing_docs)] +pub mod activity_queue; pub mod activity_sending; #[cfg(feature = "actix-web")] pub mod actix_web;