//! Passing a writer's changes on along its tree: every member gets them once, over a //! link that works, and as soon as the member before it has them. //! //! The tree is the same on every node, computed from what they share: the graph the //! member list gives (`topology::member_list_graph`: every member's neighbours, as if //! all reached all) and what each member tells its neighbours it reaches //! (`swarm::reach`). Every member's distance from the writer over that graph orders the //! members: by distance, then whether it reaches a member a hop closer, then pair key; //! a member's parent is the neighbour it reaches that comes before it, the closest to //! the writer first, then one that reaches closer, then by pair score //! (`topology::pair_score`). Coming before rules out cycles, and taking only a neighbour //! it reaches routes around links that don't work: where a writer behind NAT reaches a //! single member, that one comes first and the others take it. //! //! A sync message that passes a writer's change on says whose (`SyncEnvelope::writer`). //! Whoever applies one passes it on to its own children in that writer's tree //! (`forward`), however it got it; so does the writer, to all its neighbours, with its //! own writes. A member no member it reaches comes before (or whose parent didn't get //! the change) gets it from the announcements (`swarm::relay`), as before. //! //! The graph is computed once per member list (tens of milliseconds for thousands of //! members), the distances once per writer and graph (`MAX_WRITERS` kept). use std::collections::{BTreeMap, BTreeSet, HashMap, VecDeque}; use std::sync::Arc; use tracing::trace; use crate::locks::MutexExt; use crate::state::AppState; use crate::swarm::{reach, topology}; /// The writers whose distances are kept at most. const MAX_WRITERS: usize = 256; type Graph = BTreeMap>; /// The distances from each writer, over the graph they were computed for. #[derive(Default)] pub(crate) struct Trees { graph: Option>, /// The members by pair key, of that graph. by_key: HashMap, distances: HashMap>>, } /// The graph, and every member's distance from `writer` (by pair key) over it; None /// while the member list isn't known, or `writer` isn't in it. fn distances(state: &AppState, writer: u64) -> Option<(Arc, Arc>)> { let graph = topology::member_list_graph(state)?; let mut trees = state.trees.lock_or_recover(); if !trees .graph .as_ref() .is_some_and(|known| Arc::ptr_eq(known, &graph)) { trees.by_key = graph .keys() .map(|member| (topology::pair_key(member), member.clone())) .collect(); trees.distances.clear(); trees.graph = Some(Arc::clone(&graph)); } if let Some(known) = trees.distances.get(&writer) { return Some((graph, Arc::clone(known))); } let from = trees.by_key.get(&writer)?.clone(); let mut distance = HashMap::from([(from.clone(), 0u32)]); let mut queue = VecDeque::from([from]); while let Some(member) = queue.pop_front() { let next = distance[&member] + 1; for neighbour in graph.get(&member).into_iter().flatten() { if !distance.contains_key(neighbour) { distance.insert(neighbour.clone(), next); queue.push_back(neighbour.clone()); } } } if trees.distances.len() >= MAX_WRITERS { trees.distances.clear(); } let distance = Arc::new(distance); trees.distances.insert(writer, Arc::clone(&distance)); Some((graph, distance)) } /// Our children in `writer`'s tree: our neighbours in the graph whose parent we are, /// and whom we reach. /// /// Members are ordered by distance from the writer, then whether they reach a member a /// hop closer (as they say: else, while it hasn't the change from its own side, it is /// as good as farther), then pair key; a member's parent is the neighbour it reaches /// that comes before it, by the same, then pair score. So where the writer reaches few /// (behind NAT), those it reaches come first among the members as far from it, and the /// others take them as parents. What a member reaches we know of ourselves and our /// neighbours, as they said; of others, the graph is taken as right (if we're wrong, /// a change goes twice or by announcement). pub(crate) async fn children(state: &AppState, writer: u64) -> Vec { let Some((graph, distance)) = distances(state, writer) else { return Vec::new(); }; let me = state.local_id(); if !distance.contains_key(me) { return Vec::new(); } // Whom we and our neighbours reach, by pair key. let mut reaches: HashMap<&str, Vec> = HashMap::new(); reaches.insert(me, reach::published(state)); for member in graph.get(me).into_iter().flatten() { if let Some(peer_state) = state.get_peer_state(member).await && let Some(keys) = reach::of(state, peer_state.lock().await.reaches.as_ref()) { reaches.insert(member.as_str(), keys); } } let reached = |by: &str, member: &str| { reaches .get(by) .is_none_or(|keys| keys.contains(&topology::pair_key(member))) }; // Whether a member reaches one a hop closer to the writer (the writer: yes). let closer = |member: &str| { let d = distance[member]; d == 0 || graph.get(member).into_iter().flatten().any(|other| { distance.get(other.as_str()) == Some(&(d - 1)) && reached(member, other) }) }; let rank = |member: &str| { distance .get(member) .map(|d| (*d, !closer(member), topology::pair_key(member))) }; let Some(my_rank) = rank(me) else { return Vec::new(); }; let mut children = Vec::new(); for member in graph.get(me).into_iter().flatten() { let Some(its_rank) = rank(member) else { continue; }; if its_rank <= my_rank { continue; } let parent = graph .get(member) .into_iter() .flatten() .filter(|candidate| rank(candidate).is_some_and(|r| r < its_rank)) .filter(|candidate| reached(member, candidate)) .min_by_key(|candidate| { let (d, far, _) = rank(candidate).expect("ranked"); (d, far, topology::pair_score(candidate, member)) }); if parent.is_some_and(|parent| parent == me) && topology::reaches(state, member).await { children.push(member.clone()); } } children } /// We applied a change of `writer`'s that `from` passed on: to our children in its /// tree (but `from`), a sync message each, saying whose. Returns them. pub(crate) async fn forward(state: &AppState, writer: u64, from: &str) -> Vec { let children: Vec = children(state, writer) .await .into_iter() .filter(|child| child != from) .collect(); for child in &children { if let Some(peer_state) = state.get_peer_state(child).await { peer_state.lock().await.tree_writer = Some(writer); } state.stats.tree_forward(); state.request_sync(child); } if !children.is_empty() { trace!( writer, children = children.len(), "Passing a change on along the writer's tree" ); } children }