gitks/actor/handler.rs

use crate::actor::message::{
    ElectionRequest, ElectionResult, GitNodeMessage, NodeHealth, ROLE_PRIMARY, ROLE_REPLICA,
    RefUpdateEvent, RoleChangedEvent, RouteDecision,
};
use crate::server::GitksService;
use async_trait::async_trait;
use ractor::pg;
use ractor::{Actor, ActorProcessingErr, ActorRef, SupervisionEvent};
use std::collections::HashMap;

#[derive(Clone)]
pub struct GitNodeActor {
    pub version: String,
    pub service: GitksService,
}

impl GitNodeActor {
    pub fn init(service: GitksService) -> Self {
        GitNodeActor {
            version: env!("CARGO_PKG_VERSION").to_string(),
            service,
        }
    }
}

pub struct RepoEntry {
    pub role: String,
    pub last_commit: String,
    pub read_only: bool,
}

pub struct GitNodeArgs {
    pub storage_name: String,
    pub grpc_addr: String,
}

pub struct GitNodeState {
    storage_name: String,
    actor_name: String,
    grpc_addr: String,
    repos: HashMap<String, RepoEntry>,
    current_term: u64,
    health_failures: u32,
    is_primary: bool,
    last_known_primary_grpc: String,
}

#[async_trait]
impl Actor for GitNodeActor {
    type Msg = GitNodeMessage;
    type State = GitNodeState;
    type Arguments = GitNodeArgs;

    async fn pre_start(
        &self,
        myself: ActorRef<Self::Msg>,
        args: Self::Arguments,
    ) -> Result<Self::State, ActorProcessingErr> {
        let actor_name = format!("git_node_{}", args.storage_name);
        pg::join("gitks_nodes".to_string(), vec![myself.get_cell()]);
        pg::join_scoped(
            args.storage_name.clone(),
            "node".to_string(),
            vec![myself.get_cell()],
        );
        tracing::info!(storage_name = %args.storage_name, actor_name = %actor_name, grpc_addr = %args.grpc_addr, "GitNodeActor started");

        start_health_checker(myself.clone(), 1, 10);

        Ok(GitNodeState {
            storage_name: args.storage_name,
            actor_name,
            grpc_addr: args.grpc_addr.clone(),
            repos: HashMap::new(),
            current_term: 0,
            health_failures: 0,
            is_primary: true, // Will be refined at registration
            last_known_primary_grpc: args.grpc_addr.clone(),
        })
    }

    async fn handle(
        &self,
        myself: ActorRef<Self::Msg>,
        message: Self::Msg,
        state: &mut Self::State,
    ) -> Result<(), ActorProcessingErr> {
        match message {
            GitNodeMessage::ScanAndRegister => {
                let repos = self.service.scan_all_repo()?;
                tracing::info!(storage_name = %state.storage_name, found = repos.len(), "scanning local repositories");
                crate::metrics::set_repository_count(repos.len() as u64);
                for repo_path in repos {
                    let relative_path = repo_path
                        .strip_prefix(self.service.repo_prefix.to_string_lossy().as_ref())
                        .unwrap_or(&repo_path)
                        .trim_start_matches('/')
                        .to_string();
                    register_repo(&myself, state, relative_path);
                }
            }

            GitNodeMessage::RegisterRepository(header) => {
                register_repo(&myself, state, header.relative_path);
            }

            GitNodeMessage::RemoveRepository(header) => {
                state.repos.remove(&header.relative_path);
                tracing::info!(storage_name = %state.storage_name, relative_path = %header.relative_path, "repository route removed");
            }

            GitNodeMessage::RefUpdated(event) => {
                if let Some(entry) = state.repos.get(&event.relative_path)
                    && entry.role == ROLE_REPLICA
                {
                    let local_path = self.service.repo_prefix.join(&event.relative_path);
                    crate::actor::sync::sync_from_primary(event, local_path).await;
                }
            }

            GitNodeMessage::FindPrimary(header, reply) => {
                let entry = state.repos.get(&header.relative_path);
                let is_primary = entry.is_some_and(|e| e.role == ROLE_PRIMARY);
                reply
                    .send(build_decision(
                        state,
                        &header,
                        is_primary,
                        entry.map(|e| e.role.as_str()),
                    ))
                    .ok();
            }

            GitNodeMessage::FindReplica(header, reply) => {
                let entry = state.repos.get(&header.relative_path);
                let has = entry.is_some();
                reply
                    .send(build_decision(
                        state,
                        &header,
                        has,
                        entry.map(|e| e.role.as_str()),
                    ))
                    .ok();
            }

            GitNodeMessage::ListRepositoryPaths(reply) => {
                let paths: Vec<String> = state.repos.keys().cloned().collect();
                reply.send(paths.join("\n")).ok();
            }

            GitNodeMessage::RepositoryExists(header, reply) => {
                reply
                    .send(state.repos.contains_key(&header.relative_path))
                    .ok();
            }

            GitNodeMessage::GetNodeHealth(reply) => {
                reply
                    .send(NodeHealth {
                        storage_name: state.storage_name.clone(),
                        repo_count: state.repos.len() as u64,
                        healthy: true,
                        version: self.version.clone(),
                    })
                    .ok();
            }

            GitNodeMessage::ElectPrimary(request, reply) => {
                let accepted = should_accept_election(&request, state);
                tracing::info!(
                    candidate = %request.candidate_storage_name,
                    term = request.term,
                    current_term = state.current_term,
                    accepted = accepted,
                    "election vote"
                );
                if accepted {
                    state.current_term = request.term;
                    state.last_known_primary_grpc = request.candidate_grpc_addr.clone();
                }
                reply
                    .send(ElectionResult {
                        accepted,
                        current_term: state.current_term,
                        voter_storage_name: state.storage_name.clone(),
                        voter_role: if state.is_primary {
                            ROLE_PRIMARY
                        } else {
                            ROLE_REPLICA
                        }
                        .to_string(),
                    })
                    .ok();
            }

            GitNodeMessage::RoleChanged(event) => {
                // Empty storage_name = self-promotion from health checker
                let is_self =
                    event.storage_name.is_empty() || event.storage_name == state.storage_name;

                if is_self && event.new_role == ROLE_PRIMARY {
                    tracing::info!(
                        storage_name = %state.storage_name,
                        term = event.term,
                        "promoted to PRIMARY"
                    );
                    state.is_primary = true;
                    state.current_term = event.term;
                    state.health_failures = 0;
                    for entry in state.repos.values_mut() {
                        entry.role = ROLE_PRIMARY.to_string();
                        entry.read_only = false;
                    }
                } else if is_self && event.new_role == ROLE_REPLICA {
                    tracing::info!(
                        storage_name = %state.storage_name,
                        term = event.term,
                        "demoted to REPLICA"
                    );
                    state.is_primary = false;
                    state.current_term = event.term;
                    for entry in state.repos.values_mut() {
                        entry.role = ROLE_REPLICA.to_string();
                    }
                } else {
                    // Another node's role changed — update routing info
                    tracing::info!(
                        storage_name = %event.storage_name,
                        new_role = %event.new_role,
                        "remote node role changed"
                    );
                    state.last_known_primary_grpc = if event.new_role == ROLE_PRIMARY {
                        event.grpc_addr.clone()
                    } else {
                        state.last_known_primary_grpc.clone()
                    };
                }
            }
        }
        Ok(())
    }

    async fn handle_supervisor_evt(
        &self,
        _myself: ActorRef<Self::Msg>,
        evt: SupervisionEvent,
        _state: &mut Self::State,
    ) -> Result<(), ActorProcessingErr> {
        match evt {
            SupervisionEvent::ActorStarted(who) => {
                tracing::debug!(actor = ?who.get_id(), "child started")
            }
            SupervisionEvent::ActorTerminated(who, _, reason) => {
                tracing::warn!(actor = ?who.get_id(), reason = ?reason, "child terminated")
            }
            SupervisionEvent::ActorFailed(who, panic_msg) => {
                tracing::error!(actor = ?who.get_id(), msg = %panic_msg, "child panicked")
            }
            SupervisionEvent::ProcessGroupChanged(group) => {
                tracing::info!(group = ?group, "PG membership changed")
            }
            _ => {}
        }
        Ok(())
    }

    async fn post_stop(
        &self,
        _myself: ActorRef<Self::Msg>,
        state: &mut Self::State,
    ) -> Result<(), ActorProcessingErr> {
        tracing::info!(storage_name = %state.storage_name, "GitNodeActor stopped");
        Ok(())
    }
}

/// Determine whether to accept an election request.
fn should_accept_election(request: &ElectionRequest, state: &GitNodeState) -> bool {
    // Only accept if the term is greater than our current term
    // (prevents old/duplicate election messages)
    if request.term <= state.current_term {
        tracing::warn!(
            request_term = request.term,
            current_term = state.current_term,
            "rejecting election: term too old"
        );
        return false;
    }
    true
}

fn build_decision(
    state: &GitNodeState,
    header: &crate::pb::RepositoryHeader,
    found: bool,
    role: Option<&str>,
) -> RouteDecision {
    RouteDecision {
        found,
        storage_name: if found {
            state.storage_name.clone()
        } else {
            String::new()
        },
        relative_path: header.relative_path.clone(),
        actor_name: if found {
            state.actor_name.clone()
        } else {
            String::new()
        },
        grpc_addr: if found {
            state.grpc_addr.clone()
        } else {
            String::new()
        },
        role: role.unwrap_or("").to_string(),
    }
}

fn register_repo(
    myself: &ActorRef<GitNodeMessage>,
    state: &mut GitNodeState,
    relative_path: String,
) {
    if state.repos.contains_key(&relative_path) {
        return;
    }

    let members = ractor::pg::get_members(&"gitks_nodes".to_string());
    let my_cell = myself.get_cell();
    let other_nodes_exist = members.iter().any(|m| m != &my_cell);

    let role = if other_nodes_exist {
        ROLE_REPLICA.to_string()
    } else {
        ROLE_PRIMARY.to_string()
    };

    if role == ROLE_PRIMARY {
        state.is_primary = true;
    }

    let category = extract_category(&relative_path);
    pg::join_scoped(
        state.storage_name.clone(),
        category.to_string(),
        vec![myself.get_cell()],
    );
    state.repos.insert(
        relative_path.clone(),
        RepoEntry {
            role: role.clone(),
            last_commit: String::new(),
            read_only: false,
        },
    );
    tracing::info!(
        storage_name = %state.storage_name,
        category = %category,
        relative_path = %relative_path,
        actor_name = %state.actor_name,
        role = %role,
        "repository route registered"
    );
}

fn extract_category(relative_path: &str) -> &str {
    relative_path.split('/').next().unwrap_or("root")
}

/// Start background health checker that monitors the PRIMARY node.
/// If the PRIMARY becomes unreachable for `max_failures` consecutive checks,
/// triggers an election.
fn start_health_checker(myself: ActorRef<GitNodeMessage>, interval_secs: u64, max_failures: u32) {
    tokio::spawn(async move {
        let mut interval = tokio::time::interval(std::time::Duration::from_secs(interval_secs));
        interval.tick().await; // First tick immediate

        let mut consecutive_failures: u32 = 0;

        loop {
            interval.tick().await;

            let members = ractor::pg::get_members(&"gitks_nodes".to_string());
            let my_cell = myself.get_cell();
            let other_cells: Vec<ractor::ActorCell> =
                members.into_iter().filter(|m| m != &my_cell).collect();

            if other_cells.is_empty() {
                // No other nodes → we are the only node → ensure we are PRIMARY
                consecutive_failures = 0;
                continue;
            }

            let mut any_reachable = false;
            for cell in &other_cells {
                let actor_ref: ActorRef<GitNodeMessage> = cell.clone().into();
                match ractor::call_t!(actor_ref, GitNodeMessage::GetNodeHealth, 2000) {
                    Ok(health) if health.healthy => {
                        any_reachable = true;
                        break;
                    }
                    _ => continue,
                }
            }

            if any_reachable {
                consecutive_failures = 0;
            } else {
                consecutive_failures += 1;
                tracing::warn!(
                    consecutive_failures = consecutive_failures,
                    max_failures = max_failures,
                    "no other cluster nodes reachable"
                );

                if consecutive_failures >= max_failures {
                    tracing::error!(
                        "no other nodes reachable for {max_failures} checks, triggering self-election as PRIMARY"
                    );
                    trigger_self_election(&myself);
                    consecutive_failures = 0;
                }
            }
        }
    });
}

/// Trigger self-election: this node promotes itself to PRIMARY.
fn trigger_self_election(myself: &ActorRef<GitNodeMessage>) {
    let members = ractor::pg::get_members(&"gitks_nodes".to_string());
    let total_nodes = members.len();

    tracing::warn!(
        total_nodes = total_nodes,
        "initiating self-election as new PRIMARY"
    );

    let new_term = std::time::SystemTime::now()
        .duration_since(std::time::SystemTime::UNIX_EPOCH)
        .unwrap_or_default()
        .as_secs();

    myself
        .cast(GitNodeMessage::RoleChanged(RoleChangedEvent {
            storage_name: String::new(), // will be filled by handler from our own state
            grpc_addr: String::new(),
            new_role: ROLE_PRIMARY.to_string(),
            term: new_term,
            relative_paths: Vec::new(), // all repos
        }))
        .ok();

    broadcast_role_changed(
        myself,
        RoleChangedEvent {
            storage_name: String::new(), // handler fills
            grpc_addr: String::new(),
            new_role: ROLE_PRIMARY.to_string(),
            term: new_term,
            relative_paths: Vec::new(),
        },
    );
}

pub async fn start_node_actor(
    service: GitksService,
    storage_name: String,
    grpc_addr: String,
) -> Result<(ActorRef<GitNodeMessage>, tokio::task::JoinHandle<()>), ractor::SpawnErr> {
    let actor = GitNodeActor::init(service);
    let (actor_ref, handle) = Actor::spawn(
        Some(format!("git_node_{storage_name}")),
        actor,
        GitNodeArgs {
            storage_name,
            grpc_addr,
        },
    )
    .await?;
    actor_ref.cast(GitNodeMessage::ScanAndRegister).ok();
    Ok((actor_ref, handle))
}

pub fn get_cluster_nodes(storage_name: &str) -> Vec<ractor::ActorCell> {
    pg::get_scoped_members(&storage_name.to_string(), &"node".to_string())
}

pub fn get_category_members(storage_name: &str, category: &str) -> Vec<ractor::ActorCell> {
    pg::get_scoped_members(&storage_name.to_string(), &category.to_string())
}

pub fn route_group_for(header: &crate::pb::RepositoryHeader) -> String {
    extract_category(&header.relative_path).to_string()
}

pub fn list_all_groups() -> Vec<String> {
    pg::which_groups()
}

pub fn broadcast_ref_update(_node_actor: &ActorRef<GitNodeMessage>, event: RefUpdateEvent) {
    let members = ractor::pg::get_members(&"gitks_nodes".to_string());
    for member in members {
        let actor_ref: ActorRef<GitNodeMessage> = member.into();
        actor_ref
            .cast(GitNodeMessage::RefUpdated(event.clone()))
            .ok();
    }
}

/// Broadcast a role change event to all cluster members.
pub fn broadcast_role_changed(_actor: &ActorRef<GitNodeMessage>, event: RoleChangedEvent) {
    let members = ractor::pg::get_members(&"gitks_nodes".to_string());
    for member in members {
        let actor_ref: ActorRef<GitNodeMessage> = member.into();
        actor_ref
            .cast(GitNodeMessage::RoleChanged(event.clone()))
            .ok();
    }
}